From a0df276e976f213f64cf2a6247ba5216acae0309 Mon Sep 17 00:00:00 2001
From: Lesheng Jin <leshenj15@gmail.com>
Date: Thu, 24 Aug 2023 20:57:31 +0000
Subject: [PATCH 1/3] add whisper and test

---
 mlc_llm/models/whisper.py          | 430 +++++++++++++++++++++++++++++
 tests/python/test_model_whisper.py | 168 +++++++++++
 2 files changed, 598 insertions(+)
 create mode 100644 mlc_llm/models/whisper.py
 create mode 100644 tests/python/test_model_whisper.py

diff --git a/mlc_llm/models/whisper.py b/mlc_llm/models/whisper.py
new file mode 100644
index 0000000000..686a996cba
--- /dev/null
+++ b/mlc_llm/models/whisper.py
@@ -0,0 +1,430 @@
+import dataclasses
+from typing import Tuple, Optional
+
+import tvm
+from tvm import te, tir
+import tvm.relax.frontend.nn as nn
+from tvm.relax.frontend.nn import (
+    Embedding,
+    KVCache,
+    Linear,
+    LayerNorm,
+    Conv1D,
+    Module,
+    ModuleList,
+    Parameter,
+    Tensor,
+    tensor_expr_op,
+    permute_dims,
+    reshape,
+    squeeze,
+    matmul,
+    maximum,
+    minimum,
+    softmax,
+    gelu,
+    print_,
+)
+
+
+@dataclasses.dataclass
+class WhisperConfig:
+    def __init__(
+        self,
+        dtype="float32",
+        vocab_size=51865,
+        num_mel_bins=80,
+        encoder_layers=6,
+        encoder_attention_heads=4,
+        decoder_layers=6,
+        decoder_attention_heads=4,
+        decoder_ffn_dim=1536,
+        encoder_ffn_dim=1536,
+        decoder_start_token_id=50257,
+        d_model=256,
+        max_source_positions=1500,
+        max_target_positions=448,
+        pad_token_id=50256,
+        bos_token_id=50257,
+        eos_token_id=50256,
+        suppress_tokens=None,
+        begin_suppress_tokens=[220, 50256],
+        forced_decoder_ids=None,
+        **kwargs,
+    ):
+        self.dtype = dtype
+        self.vocab_size = vocab_size
+        self.num_mel_bins = num_mel_bins
+        self.d_model = d_model
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.num_hidden_layers = encoder_layers
+        self.max_source_positions = max_source_positions
+        self.max_target_positions = max_target_positions
+        self.pad_token_id = pad_token_id
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.decoder_start_token_id = decoder_start_token_id
+        self.suppress_tokens = suppress_tokens
+        self.begin_suppress_tokens = begin_suppress_tokens
+        self.forced_decoder_ids = forced_decoder_ids
+        self.kwargs = kwargs
+
+
+class WhisperPositionalEmbedding(Module):
+    def __init__(self, max_seq_len: int, embed_dim: int, dtype: str):
+        super().__init__()
+        self.max_seq_len = max_seq_len
+        self.embed_dim = embed_dim
+        self.weight = Parameter((max_seq_len, embed_dim), dtype=dtype)
+
+    def forward(self, x: Tensor, offset: tir.Var):
+        def te_op(x: te.Tensor, embed: te.Tensor, offset: tir.Var):
+            def compute(i: tir.Var, j: tir.Var, k: tir.Var):
+                return embed[offset + j, k]
+
+            return tvm.te.compute([*x.shape, embed.shape[-1]], compute, name="position_embedding")
+
+        pos_embed = tensor_expr_op(te_op, "position_embedding", args=[x, self.weight, offset])
+        return pos_embed
+
+
+class WhisperAttention(Module):
+    def __init__(
+        self, embed_dim: int, num_heads: int, kv_cache_len: int, dtype: str, bias: bool = True
+    ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        self.head_dim = self.embed_dim // self.num_heads
+
+        if (self.head_dim * num_heads) != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
+                f" and `num_heads`: {num_heads})."
+            )
+        self.scaling = self.head_dim**-0.5
+        self.k_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=False)
+        self.v_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
+        self.q_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
+        self.out_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
+        if kv_cache_len > 0:
+            self.k_cache = KVCache(kv_cache_len, [self.num_heads, self.head_dim])
+            self.v_cache = KVCache(kv_cache_len, [self.num_heads, self.head_dim])
+
+    def forward(
+        self,
+        hidden_states: Tensor,
+        key_value_states: Optional[Tensor] = None,
+        cached_cross_attn_states: Optional[Tuple[Tensor]] = None,
+        attention_mask: Optional[Tensor] = None,
+        total_seq_len: Optional[tir.Var] = None,
+    ) -> Tuple[Tensor, Optional[Tuple[Tensor]]]:
+        is_cross_attention = key_value_states is not None or cached_cross_attn_states is not None
+
+        h, d = self.num_heads, self.head_dim
+        bsz, q_len, _ = hidden_states.shape
+        assert bsz == 1, "Only support batch size 1 at this moment."
+
+        q = reshape(self.q_proj(hidden_states) * self.scaling, (bsz, q_len, h, d))
+
+        if is_cross_attention:
+            # cross attention
+            if cached_cross_attn_states is None:
+                # no cache, cross attentions
+                kv_len = key_value_states.shape[1]
+                k = reshape(self.k_proj(key_value_states), (bsz, kv_len, h, d))
+                v = reshape(self.v_proj(key_value_states), (bsz, kv_len, h, d))
+                cached_kv = (k, v)
+            else:
+                # reuse cached k,v, cross_attentions
+                k, v = cached_cross_attn_states
+        else:
+            # self attention
+            k = reshape(self.k_proj(hidden_states), (bsz, q_len, h, d))
+            v = reshape(self.v_proj(hidden_states), (bsz, q_len, h, d))
+
+            if total_seq_len is not None:
+                # reuse cached k, v, self_attention
+                self.k_cache.append(squeeze(k, axis=0))
+                self.v_cache.append(squeeze(v, axis=0))
+                k = reshape(self.k_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
+                v = reshape(self.v_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
+            else:
+                # encode self attention, no cache
+                ...
+
+        q = permute_dims(q, [0, 2, 1, 3])  # [b, h, q_len, d]
+        k = permute_dims(k, [0, 2, 1, 3])  # [b, h, q_len, d]
+        v = permute_dims(v, [0, 2, 1, 3])  # [b, h, q_len, d]
+
+        attn_weights = matmul(q, (permute_dims(k, [0, 1, 3, 2])))  # [b, h, q_len, q_len]
+
+        if attention_mask is not None:
+            attn_weights = attn_weights + attention_mask
+
+        dtype = attn_weights.dtype
+        attn_weights = attn_weights.maximum(tir.min_value(dtype))
+        attn_weights = attn_weights.minimum(tir.max_value(dtype))
+        if dtype == "float32":
+            attn_weights = softmax(attn_weights, axis=-1)
+        else:
+            attn_weights = softmax(attn_weights.astype("float32"), axis=-1).astype(dtype)
+        attn_output = matmul(attn_weights, v)  # [b, h, q_len, d]
+
+        attn_output = permute_dims(attn_output, [0, 2, 1, 3])  # [b, q_len, h, d]
+        attn_output = reshape(attn_output, (bsz, q_len, self.embed_dim))  # [b, q_len, h * d]
+
+        attn_output = self.out_proj(attn_output)
+
+        if is_cross_attention and cached_cross_attn_states is None:
+            return attn_output, cached_kv
+        else:
+            return attn_output, None
+
+
+class EncoderLayer(Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.self_attn = WhisperAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.encoder_attention_heads,
+            kv_cache_len=0,  # no need for kv_cache
+            dtype=config.dtype,
+        )
+        self.self_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
+        self.fc1 = Linear(self.embed_dim, config.encoder_ffn_dim, dtype=config.dtype)
+        self.fc2 = Linear(config.encoder_ffn_dim, self.embed_dim, dtype=config.dtype)
+        self.final_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
+
+    def forward(self, hidden_states: Tensor, attention_mask: Optional[Tensor] = None) -> Tensor:
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            key_value_states=None,
+            cached_cross_attn_states=None,
+            attention_mask=attention_mask,
+            total_seq_len=None,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = gelu(self.fc1(hidden_states))
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = residual + hidden_states
+        hidden_states = hidden_states.maximum(tir.min_value(hidden_states.dtype))
+        hidden_states = hidden_states.minimum(tir.max_value(hidden_states.dtype))
+
+        return hidden_states
+
+
+class DecoderLayer(Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+
+        self.self_attn = WhisperAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.decoder_attention_heads,
+            kv_cache_len=100,  # TODO
+            dtype=config.dtype,
+        )
+
+        self.self_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
+        self.encoder_attn = WhisperAttention(
+            self.embed_dim,
+            config.decoder_attention_heads,
+            kv_cache_len=100,  # TODO
+            dtype=config.dtype,
+        )
+        self.encoder_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
+        self.fc1 = Linear(self.embed_dim, config.decoder_ffn_dim, dtype=config.dtype)
+        self.fc2 = Linear(config.decoder_ffn_dim, self.embed_dim, dtype=config.dtype)
+        self.final_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
+
+    def forward(
+        self,
+        hidden_states: Tensor,
+        encoder_hidden_states: Tensor,
+        cached_encoder_hidden_states: Tensor,
+        total_seq_len: tir.Var,
+        attention_mask: Optional[Tensor] = None,
+        encoder_attention_mask: Optional[Tensor] = None,
+    ) -> Tuple[Tensor, Optional[Tuple[Tensor]]]:
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+
+        # Self Attention
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            total_seq_len=total_seq_len,
+            key_value_states=None,
+            cached_cross_attn_states=None,
+            attention_mask=attention_mask,
+        )
+        hidden_states = residual + hidden_states
+
+        # Cross-Attention Block
+        residual = hidden_states
+        hidden_states = self.encoder_attn_layer_norm(hidden_states)
+        hidden_states, cross_attn_key_value = self.encoder_attn(
+            hidden_states=hidden_states,
+            total_seq_len=total_seq_len,
+            key_value_states=encoder_hidden_states,
+            cached_cross_attn_states=cached_encoder_hidden_states,
+            attention_mask=encoder_attention_mask,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = gelu(self.fc1(hidden_states))
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = residual + hidden_states
+
+        if cached_encoder_hidden_states is None:
+            return hidden_states, cross_attn_key_value
+        else:
+            return hidden_states, None
+
+
+class WhisperEncoder(Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+
+        embed_dim = config.d_model
+        self.num_mel_bins = config.num_mel_bins
+        self.padding_idx = config.pad_token_id
+        self.max_source_positions = config.max_source_positions
+        self.embed_scale = 1.0
+
+        self.conv1 = Conv1D(
+            self.num_mel_bins, embed_dim, kernel_size=3, padding=1, dtype=config.dtype
+        )
+        self.conv2 = Conv1D(
+            embed_dim, embed_dim, kernel_size=3, stride=2, padding=1, dtype=config.dtype
+        )
+
+        self.embed_positions = Embedding(self.max_source_positions, embed_dim, dtype=config.dtype)
+
+        self.layers = ModuleList([EncoderLayer(config) for _ in range(config.encoder_layers)])
+        self.layer_norm = LayerNorm(config.d_model, dtype=config.dtype)
+
+    def forward(self, input_features: Tensor) -> Tensor:
+        inputs_embeds = gelu(self.conv1(input_features))
+        inputs_embeds = gelu(self.conv2(inputs_embeds))
+
+        inputs_embeds = permute_dims(inputs_embeds, [0, 2, 1])
+        embed_pos = self.embed_positions.weight
+
+        hidden_states = inputs_embeds + embed_pos
+
+        for idx, encoder_layer in enumerate(self.layers):
+            hidden_states = encoder_layer(hidden_states)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        return hidden_states
+
+
+class WhisperDecoder(Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+
+        self.max_target_positions = config.max_target_positions
+        self.max_source_positions = config.max_source_positions
+        self.embed_scale = 1.0
+        self.embed_tokens = Embedding(config.vocab_size, config.d_model, dtype=config.dtype)
+        self.embed_positions = WhisperPositionalEmbedding(
+            self.max_target_positions, config.d_model, dtype=config.dtype
+        )
+
+        self.layers = ModuleList([DecoderLayer(config) for _ in range(config.decoder_layers)])
+
+        self.layer_norm = LayerNorm(config.d_model, dtype=config.dtype)
+
+    def forward(
+        self,
+        input_ids: Tensor,
+        total_seq_len: Optional[tir.Var] = None,
+        encoder_hidden_states: Optional[Tensor] = None,
+        cached_encoder_key_value: Optional[Tuple[Tuple[Tensor]]] = None,
+        attention_mask: Optional[Tensor] = None,
+    ):
+        # total_seq_len = Length of generated tokens
+        input_embeds = self.embed_tokens(input_ids)
+        past_seq_len = total_seq_len - 1
+        position_embeds = self.embed_positions(input_ids, offset=past_seq_len)
+
+        hidden_states = input_embeds + position_embeds
+
+        all_encoder_key_value = ()
+        for idx, decoder_layer in enumerate(self.layers):
+            ith_cached_encoder_key_value = (
+                cached_encoder_key_value[idx] if cached_encoder_key_value is not None else None
+            )
+            hidden_states, encoder_key_value = decoder_layer(
+                hidden_states=hidden_states,
+                total_seq_len=total_seq_len,
+                encoder_hidden_states=encoder_hidden_states,
+                cached_encoder_hidden_states=ith_cached_encoder_key_value,
+                attention_mask=attention_mask,
+            )
+            if cached_encoder_key_value is None:
+                all_encoder_key_value += (encoder_key_value,)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        if cached_encoder_key_value is None:
+            return hidden_states, all_encoder_key_value
+        else:
+            return hidden_states, None
+
+
+class WhisperModel(Module):
+    def __init__(self, config: WhisperConfig):
+        self.encoder = WhisperEncoder(config)
+        self.decoder = WhisperDecoder(config)
+
+
+class WhisperForConditionalGeneration(Module):
+    def __init__(self, config: WhisperConfig):
+        self.model = WhisperModel(config)
+        self.proj_out = Linear(config.d_model, config.vocab_size, bias=False, dtype=config.dtype)
+
+    def encode(self, input_features: Tensor) -> Tensor:
+        return self.model.encoder(input_features)
+
+    def decode(
+        self, input_ids: Tensor, total_seq_len: int, encoder_hidden_states: Tensor
+    ) -> Tuple[Tensor, Tuple[Tuple[Tensor]]]:
+        hidden_states, all_encoder_key_value = self.model.decoder.forward(
+            input_ids=input_ids,
+            total_seq_len=total_seq_len,
+            encoder_hidden_states=encoder_hidden_states,
+            cached_encoder_key_value=None,
+            attention_mask=None,
+        )
+        lm_logits = self.proj_out(hidden_states)
+        return lm_logits, all_encoder_key_value
+
+    def prefill(
+        self, input_ids: Tensor, total_seq_len: int, cached_encoder_key_value: Tuple[Tuple[Tensor]]
+    ) -> Tensor:
+        hidden_states, _ = self.model.decoder.forward(
+            input_ids=input_ids,
+            total_seq_len=total_seq_len,
+            encoder_hidden_states=None,
+            cached_encoder_key_value=cached_encoder_key_value,
+            attention_mask=None,
+        )
+        lm_logits = self.proj_out(hidden_states)
+        return lm_logits
diff --git a/tests/python/test_model_whisper.py b/tests/python/test_model_whisper.py
new file mode 100644
index 0000000000..40515954a9
--- /dev/null
+++ b/tests/python/test_model_whisper.py
@@ -0,0 +1,168 @@
+# pylint: disable=invalid-name,missing-docstring
+from typing import Tuple, Any
+
+import torch
+from transformers import WhisperForConditionalGeneration as hf_Whisper
+from transformers import WhisperProcessor
+from datasets import load_dataset
+import torch
+
+import tvm
+from tvm.relax.frontend.nn import spec
+
+from mlc_llm.models.whisper import WhisperConfig, WhisperForConditionalGeneration
+
+
+def load_whisper_from_hf() -> Tuple[hf_Whisper, WhisperProcessor]:
+    processor = WhisperProcessor.from_pretrained("openai/whisper-medium")
+    hf_model = hf_Whisper.from_pretrained("openai/whisper-medium")
+    hf_model = hf_model.eval().to("cuda")
+    return hf_model, processor
+
+
+def load_data(processor: WhisperProcessor, test_idx: int) -> Tuple[torch.Tensor, str]:
+    ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
+    sample = ds[test_idx]["audio"]
+    text = ds[test_idx]["text"]
+    input_features = processor(
+        sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt"
+    ).input_features.to("cuda")
+
+    return input_features, text
+
+
+def pipe(model: Any, config: WhisperConfig, input_features) -> torch.Tensor:
+    # encode
+    encode_output = model["encode"](input_features)
+
+    # decode start token
+    input_ids = torch.tensor([[config.decoder_start_token_id]], dtype=torch.int32).to("cuda")
+    generated_tokens = [config.decoder_start_token_id]
+
+    while True:
+        if len(generated_tokens) == 1:
+            outputs, encode_kv_cache = model["decode"](
+                input_ids, len(generated_tokens), encode_output
+            )
+        else:
+            outputs = model["prefill"](input_ids, len(generated_tokens), encode_kv_cache)
+
+        outputs_logits = outputs
+        next_token_logits = outputs_logits[:, 0, :]
+
+        # suppress tokens
+        next_tokens_scores = next_token_logits
+        next_tokens_scores[:, config.suppress_tokens] = -float("inf")
+
+        # suppress tokens at begin
+        if len(generated_tokens) == 1 + config.forced_decoder_ids[-1][0]:
+            next_tokens_scores[:, config.begin_suppress_tokens] = -float("inf")
+
+        # force tokens at sepcific position
+        generation_idx = len(generated_tokens)
+        current_token = dict(config.forced_decoder_ids).get(generation_idx, None)
+        if current_token is not None:
+            next_tokens_scores[:, :] = -float("inf")
+            next_tokens_scores[:, current_token] = 0
+
+        # argmax
+        next_token = torch.argmax(next_tokens_scores, dim=-1)[0]
+        input_ids[0][0] = next_token
+
+        generated_tokens.append(next_token)
+
+        # stop when we meet eos_token_id or exceed the maximum length
+        if (
+            next_token == config.eos_token_id
+            or len(generated_tokens) == config.max_target_positions
+        ):
+            break
+
+    return generated_tokens
+
+
+def main():
+    # Set the device and target
+    dev = tvm.cuda()
+    target = tvm.target.Target(
+        {
+            "kind": "cuda",
+            "max_shared_memory_per_block": dev.max_shared_memory_per_block,
+            "max_threads_per_block": dev.max_threads_per_block,
+            "thread_warp_size": dev.warp_size,
+            "registers_per_block": 65536,
+            "arch": "sm_" + tvm.cuda().compute_version.replace(".", ""),
+        }
+    )
+
+    # load model from transformers
+    hf_model, processor = load_whisper_from_hf()
+
+    # Define the model config
+    config = WhisperConfig(**hf_model.config.to_dict())
+    bsz, encode_input_ndim = 1, 16000 * 30 // 160
+
+    # Define the model
+    model = WhisperForConditionalGeneration(config=config)
+
+    mod_spec = {
+        "encode": {
+            "input_features": spec.Tensor([bsz, config.num_mel_bins, encode_input_ndim], "float32"),
+        },
+        "decode": {
+            "input_ids": spec.Tensor([bsz, "seq_len"], "int32"),
+            "total_seq_len": int,
+            "encoder_hidden_states": spec.Tensor(
+                [bsz, config.max_source_positions, config.d_model], "float32"
+            ),
+        },
+        "prefill": {
+            "input_ids": spec.Tensor([bsz, 1], "int32"),
+            "total_seq_len": int,
+            "cached_encoder_key_value": tuple(
+                tuple(
+                    spec.Tensor(
+                        [
+                            1,
+                            config.max_source_positions,
+                            config.decoder_attention_heads,
+                            config.d_model // config.decoder_attention_heads,
+                        ],
+                        "float32",
+                    )
+                    for i2 in range(2)
+                )
+                for i1 in range(config.num_hidden_layers)
+            ),
+        },
+    }
+
+    # Usercase1, export it to TVM's IRModule, use `mod.show()` to print the IRModule
+    mod, _ = model.export_tvm(spec=mod_spec)
+
+    # Usercase2, JIT compile a model
+    for name, param in model.state_dict().items():
+        param.data = hf_model.state_dict()[name]
+
+    model = model.jit(spec=mod_spec, target=target, device="cuda", out_format="torch", debug=True)
+
+    # Test on librispeech_asr_dummy
+    input_features, text = load_data(processor, test_idx=0)
+    generated_tokens = pipe(model, config, input_features)
+
+    # compare with hf whisper output
+    hf_predicted_ids = hf_model.generate(input_features).to("cpu")
+    assert torch.equal(torch.tensor([generated_tokens], dtype=torch.long), hf_predicted_ids)
+
+    # decode token ids to text
+    output = processor.decode(generated_tokens, skip_special_tokens=True)
+    assert (
+        output
+        == " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel."
+    )
+
+    print("Transcription:\n", output)
+
+
+if __name__ == "__main__":
+    main()

From 860632763ac25e06a59cce18c251e464cafb8673 Mon Sep 17 00:00:00 2001
From: Lesheng Jin <leshenj15@gmail.com>
Date: Mon, 5 Feb 2024 20:57:50 +0000
Subject: [PATCH 2/3] fix

---
 mlc_llm/models/whisper.py | 430 --------------------------------------
 1 file changed, 430 deletions(-)
 delete mode 100644 mlc_llm/models/whisper.py

diff --git a/mlc_llm/models/whisper.py b/mlc_llm/models/whisper.py
deleted file mode 100644
index 686a996cba..0000000000
--- a/mlc_llm/models/whisper.py
+++ /dev/null
@@ -1,430 +0,0 @@
-import dataclasses
-from typing import Tuple, Optional
-
-import tvm
-from tvm import te, tir
-import tvm.relax.frontend.nn as nn
-from tvm.relax.frontend.nn import (
-    Embedding,
-    KVCache,
-    Linear,
-    LayerNorm,
-    Conv1D,
-    Module,
-    ModuleList,
-    Parameter,
-    Tensor,
-    tensor_expr_op,
-    permute_dims,
-    reshape,
-    squeeze,
-    matmul,
-    maximum,
-    minimum,
-    softmax,
-    gelu,
-    print_,
-)
-
-
-@dataclasses.dataclass
-class WhisperConfig:
-    def __init__(
-        self,
-        dtype="float32",
-        vocab_size=51865,
-        num_mel_bins=80,
-        encoder_layers=6,
-        encoder_attention_heads=4,
-        decoder_layers=6,
-        decoder_attention_heads=4,
-        decoder_ffn_dim=1536,
-        encoder_ffn_dim=1536,
-        decoder_start_token_id=50257,
-        d_model=256,
-        max_source_positions=1500,
-        max_target_positions=448,
-        pad_token_id=50256,
-        bos_token_id=50257,
-        eos_token_id=50256,
-        suppress_tokens=None,
-        begin_suppress_tokens=[220, 50256],
-        forced_decoder_ids=None,
-        **kwargs,
-    ):
-        self.dtype = dtype
-        self.vocab_size = vocab_size
-        self.num_mel_bins = num_mel_bins
-        self.d_model = d_model
-        self.encoder_layers = encoder_layers
-        self.encoder_attention_heads = encoder_attention_heads
-        self.decoder_layers = decoder_layers
-        self.decoder_attention_heads = decoder_attention_heads
-        self.decoder_ffn_dim = decoder_ffn_dim
-        self.encoder_ffn_dim = encoder_ffn_dim
-        self.num_hidden_layers = encoder_layers
-        self.max_source_positions = max_source_positions
-        self.max_target_positions = max_target_positions
-        self.pad_token_id = pad_token_id
-        self.bos_token_id = bos_token_id
-        self.eos_token_id = eos_token_id
-        self.decoder_start_token_id = decoder_start_token_id
-        self.suppress_tokens = suppress_tokens
-        self.begin_suppress_tokens = begin_suppress_tokens
-        self.forced_decoder_ids = forced_decoder_ids
-        self.kwargs = kwargs
-
-
-class WhisperPositionalEmbedding(Module):
-    def __init__(self, max_seq_len: int, embed_dim: int, dtype: str):
-        super().__init__()
-        self.max_seq_len = max_seq_len
-        self.embed_dim = embed_dim
-        self.weight = Parameter((max_seq_len, embed_dim), dtype=dtype)
-
-    def forward(self, x: Tensor, offset: tir.Var):
-        def te_op(x: te.Tensor, embed: te.Tensor, offset: tir.Var):
-            def compute(i: tir.Var, j: tir.Var, k: tir.Var):
-                return embed[offset + j, k]
-
-            return tvm.te.compute([*x.shape, embed.shape[-1]], compute, name="position_embedding")
-
-        pos_embed = tensor_expr_op(te_op, "position_embedding", args=[x, self.weight, offset])
-        return pos_embed
-
-
-class WhisperAttention(Module):
-    def __init__(
-        self, embed_dim: int, num_heads: int, kv_cache_len: int, dtype: str, bias: bool = True
-    ):
-        super().__init__()
-        self.embed_dim = embed_dim
-        self.num_heads = num_heads
-        self.head_dim = self.embed_dim // self.num_heads
-
-        if (self.head_dim * num_heads) != self.embed_dim:
-            raise ValueError(
-                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
-                f" and `num_heads`: {num_heads})."
-            )
-        self.scaling = self.head_dim**-0.5
-        self.k_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=False)
-        self.v_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
-        self.q_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
-        self.out_proj = Linear(self.embed_dim, self.embed_dim, dtype=dtype, bias=bias)
-        if kv_cache_len > 0:
-            self.k_cache = KVCache(kv_cache_len, [self.num_heads, self.head_dim])
-            self.v_cache = KVCache(kv_cache_len, [self.num_heads, self.head_dim])
-
-    def forward(
-        self,
-        hidden_states: Tensor,
-        key_value_states: Optional[Tensor] = None,
-        cached_cross_attn_states: Optional[Tuple[Tensor]] = None,
-        attention_mask: Optional[Tensor] = None,
-        total_seq_len: Optional[tir.Var] = None,
-    ) -> Tuple[Tensor, Optional[Tuple[Tensor]]]:
-        is_cross_attention = key_value_states is not None or cached_cross_attn_states is not None
-
-        h, d = self.num_heads, self.head_dim
-        bsz, q_len, _ = hidden_states.shape
-        assert bsz == 1, "Only support batch size 1 at this moment."
-
-        q = reshape(self.q_proj(hidden_states) * self.scaling, (bsz, q_len, h, d))
-
-        if is_cross_attention:
-            # cross attention
-            if cached_cross_attn_states is None:
-                # no cache, cross attentions
-                kv_len = key_value_states.shape[1]
-                k = reshape(self.k_proj(key_value_states), (bsz, kv_len, h, d))
-                v = reshape(self.v_proj(key_value_states), (bsz, kv_len, h, d))
-                cached_kv = (k, v)
-            else:
-                # reuse cached k,v, cross_attentions
-                k, v = cached_cross_attn_states
-        else:
-            # self attention
-            k = reshape(self.k_proj(hidden_states), (bsz, q_len, h, d))
-            v = reshape(self.v_proj(hidden_states), (bsz, q_len, h, d))
-
-            if total_seq_len is not None:
-                # reuse cached k, v, self_attention
-                self.k_cache.append(squeeze(k, axis=0))
-                self.v_cache.append(squeeze(v, axis=0))
-                k = reshape(self.k_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
-                v = reshape(self.v_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
-            else:
-                # encode self attention, no cache
-                ...
-
-        q = permute_dims(q, [0, 2, 1, 3])  # [b, h, q_len, d]
-        k = permute_dims(k, [0, 2, 1, 3])  # [b, h, q_len, d]
-        v = permute_dims(v, [0, 2, 1, 3])  # [b, h, q_len, d]
-
-        attn_weights = matmul(q, (permute_dims(k, [0, 1, 3, 2])))  # [b, h, q_len, q_len]
-
-        if attention_mask is not None:
-            attn_weights = attn_weights + attention_mask
-
-        dtype = attn_weights.dtype
-        attn_weights = attn_weights.maximum(tir.min_value(dtype))
-        attn_weights = attn_weights.minimum(tir.max_value(dtype))
-        if dtype == "float32":
-            attn_weights = softmax(attn_weights, axis=-1)
-        else:
-            attn_weights = softmax(attn_weights.astype("float32"), axis=-1).astype(dtype)
-        attn_output = matmul(attn_weights, v)  # [b, h, q_len, d]
-
-        attn_output = permute_dims(attn_output, [0, 2, 1, 3])  # [b, q_len, h, d]
-        attn_output = reshape(attn_output, (bsz, q_len, self.embed_dim))  # [b, q_len, h * d]
-
-        attn_output = self.out_proj(attn_output)
-
-        if is_cross_attention and cached_cross_attn_states is None:
-            return attn_output, cached_kv
-        else:
-            return attn_output, None
-
-
-class EncoderLayer(Module):
-    def __init__(self, config: WhisperConfig):
-        super().__init__()
-        self.embed_dim = config.d_model
-        self.self_attn = WhisperAttention(
-            embed_dim=self.embed_dim,
-            num_heads=config.encoder_attention_heads,
-            kv_cache_len=0,  # no need for kv_cache
-            dtype=config.dtype,
-        )
-        self.self_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
-        self.fc1 = Linear(self.embed_dim, config.encoder_ffn_dim, dtype=config.dtype)
-        self.fc2 = Linear(config.encoder_ffn_dim, self.embed_dim, dtype=config.dtype)
-        self.final_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
-
-    def forward(self, hidden_states: Tensor, attention_mask: Optional[Tensor] = None) -> Tensor:
-        residual = hidden_states
-        hidden_states = self.self_attn_layer_norm(hidden_states)
-        hidden_states, _ = self.self_attn(
-            hidden_states=hidden_states,
-            key_value_states=None,
-            cached_cross_attn_states=None,
-            attention_mask=attention_mask,
-            total_seq_len=None,
-        )
-        hidden_states = residual + hidden_states
-
-        residual = hidden_states
-        hidden_states = self.final_layer_norm(hidden_states)
-        hidden_states = gelu(self.fc1(hidden_states))
-        hidden_states = self.fc2(hidden_states)
-        hidden_states = residual + hidden_states
-        hidden_states = hidden_states.maximum(tir.min_value(hidden_states.dtype))
-        hidden_states = hidden_states.minimum(tir.max_value(hidden_states.dtype))
-
-        return hidden_states
-
-
-class DecoderLayer(Module):
-    def __init__(self, config: WhisperConfig):
-        super().__init__()
-        self.embed_dim = config.d_model
-
-        self.self_attn = WhisperAttention(
-            embed_dim=self.embed_dim,
-            num_heads=config.decoder_attention_heads,
-            kv_cache_len=100,  # TODO
-            dtype=config.dtype,
-        )
-
-        self.self_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
-        self.encoder_attn = WhisperAttention(
-            self.embed_dim,
-            config.decoder_attention_heads,
-            kv_cache_len=100,  # TODO
-            dtype=config.dtype,
-        )
-        self.encoder_attn_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
-        self.fc1 = Linear(self.embed_dim, config.decoder_ffn_dim, dtype=config.dtype)
-        self.fc2 = Linear(config.decoder_ffn_dim, self.embed_dim, dtype=config.dtype)
-        self.final_layer_norm = LayerNorm(self.embed_dim, dtype=config.dtype)
-
-    def forward(
-        self,
-        hidden_states: Tensor,
-        encoder_hidden_states: Tensor,
-        cached_encoder_hidden_states: Tensor,
-        total_seq_len: tir.Var,
-        attention_mask: Optional[Tensor] = None,
-        encoder_attention_mask: Optional[Tensor] = None,
-    ) -> Tuple[Tensor, Optional[Tuple[Tensor]]]:
-        residual = hidden_states
-        hidden_states = self.self_attn_layer_norm(hidden_states)
-
-        # Self Attention
-        hidden_states, _ = self.self_attn(
-            hidden_states=hidden_states,
-            total_seq_len=total_seq_len,
-            key_value_states=None,
-            cached_cross_attn_states=None,
-            attention_mask=attention_mask,
-        )
-        hidden_states = residual + hidden_states
-
-        # Cross-Attention Block
-        residual = hidden_states
-        hidden_states = self.encoder_attn_layer_norm(hidden_states)
-        hidden_states, cross_attn_key_value = self.encoder_attn(
-            hidden_states=hidden_states,
-            total_seq_len=total_seq_len,
-            key_value_states=encoder_hidden_states,
-            cached_cross_attn_states=cached_encoder_hidden_states,
-            attention_mask=encoder_attention_mask,
-        )
-        hidden_states = residual + hidden_states
-
-        # Fully Connected
-        residual = hidden_states
-        hidden_states = self.final_layer_norm(hidden_states)
-        hidden_states = gelu(self.fc1(hidden_states))
-        hidden_states = self.fc2(hidden_states)
-        hidden_states = residual + hidden_states
-
-        if cached_encoder_hidden_states is None:
-            return hidden_states, cross_attn_key_value
-        else:
-            return hidden_states, None
-
-
-class WhisperEncoder(Module):
-    def __init__(self, config: WhisperConfig):
-        super().__init__()
-
-        embed_dim = config.d_model
-        self.num_mel_bins = config.num_mel_bins
-        self.padding_idx = config.pad_token_id
-        self.max_source_positions = config.max_source_positions
-        self.embed_scale = 1.0
-
-        self.conv1 = Conv1D(
-            self.num_mel_bins, embed_dim, kernel_size=3, padding=1, dtype=config.dtype
-        )
-        self.conv2 = Conv1D(
-            embed_dim, embed_dim, kernel_size=3, stride=2, padding=1, dtype=config.dtype
-        )
-
-        self.embed_positions = Embedding(self.max_source_positions, embed_dim, dtype=config.dtype)
-
-        self.layers = ModuleList([EncoderLayer(config) for _ in range(config.encoder_layers)])
-        self.layer_norm = LayerNorm(config.d_model, dtype=config.dtype)
-
-    def forward(self, input_features: Tensor) -> Tensor:
-        inputs_embeds = gelu(self.conv1(input_features))
-        inputs_embeds = gelu(self.conv2(inputs_embeds))
-
-        inputs_embeds = permute_dims(inputs_embeds, [0, 2, 1])
-        embed_pos = self.embed_positions.weight
-
-        hidden_states = inputs_embeds + embed_pos
-
-        for idx, encoder_layer in enumerate(self.layers):
-            hidden_states = encoder_layer(hidden_states)
-
-        hidden_states = self.layer_norm(hidden_states)
-
-        return hidden_states
-
-
-class WhisperDecoder(Module):
-    def __init__(self, config: WhisperConfig):
-        super().__init__()
-
-        self.max_target_positions = config.max_target_positions
-        self.max_source_positions = config.max_source_positions
-        self.embed_scale = 1.0
-        self.embed_tokens = Embedding(config.vocab_size, config.d_model, dtype=config.dtype)
-        self.embed_positions = WhisperPositionalEmbedding(
-            self.max_target_positions, config.d_model, dtype=config.dtype
-        )
-
-        self.layers = ModuleList([DecoderLayer(config) for _ in range(config.decoder_layers)])
-
-        self.layer_norm = LayerNorm(config.d_model, dtype=config.dtype)
-
-    def forward(
-        self,
-        input_ids: Tensor,
-        total_seq_len: Optional[tir.Var] = None,
-        encoder_hidden_states: Optional[Tensor] = None,
-        cached_encoder_key_value: Optional[Tuple[Tuple[Tensor]]] = None,
-        attention_mask: Optional[Tensor] = None,
-    ):
-        # total_seq_len = Length of generated tokens
-        input_embeds = self.embed_tokens(input_ids)
-        past_seq_len = total_seq_len - 1
-        position_embeds = self.embed_positions(input_ids, offset=past_seq_len)
-
-        hidden_states = input_embeds + position_embeds
-
-        all_encoder_key_value = ()
-        for idx, decoder_layer in enumerate(self.layers):
-            ith_cached_encoder_key_value = (
-                cached_encoder_key_value[idx] if cached_encoder_key_value is not None else None
-            )
-            hidden_states, encoder_key_value = decoder_layer(
-                hidden_states=hidden_states,
-                total_seq_len=total_seq_len,
-                encoder_hidden_states=encoder_hidden_states,
-                cached_encoder_hidden_states=ith_cached_encoder_key_value,
-                attention_mask=attention_mask,
-            )
-            if cached_encoder_key_value is None:
-                all_encoder_key_value += (encoder_key_value,)
-
-        hidden_states = self.layer_norm(hidden_states)
-
-        if cached_encoder_key_value is None:
-            return hidden_states, all_encoder_key_value
-        else:
-            return hidden_states, None
-
-
-class WhisperModel(Module):
-    def __init__(self, config: WhisperConfig):
-        self.encoder = WhisperEncoder(config)
-        self.decoder = WhisperDecoder(config)
-
-
-class WhisperForConditionalGeneration(Module):
-    def __init__(self, config: WhisperConfig):
-        self.model = WhisperModel(config)
-        self.proj_out = Linear(config.d_model, config.vocab_size, bias=False, dtype=config.dtype)
-
-    def encode(self, input_features: Tensor) -> Tensor:
-        return self.model.encoder(input_features)
-
-    def decode(
-        self, input_ids: Tensor, total_seq_len: int, encoder_hidden_states: Tensor
-    ) -> Tuple[Tensor, Tuple[Tuple[Tensor]]]:
-        hidden_states, all_encoder_key_value = self.model.decoder.forward(
-            input_ids=input_ids,
-            total_seq_len=total_seq_len,
-            encoder_hidden_states=encoder_hidden_states,
-            cached_encoder_key_value=None,
-            attention_mask=None,
-        )
-        lm_logits = self.proj_out(hidden_states)
-        return lm_logits, all_encoder_key_value
-
-    def prefill(
-        self, input_ids: Tensor, total_seq_len: int, cached_encoder_key_value: Tuple[Tuple[Tensor]]
-    ) -> Tensor:
-        hidden_states, _ = self.model.decoder.forward(
-            input_ids=input_ids,
-            total_seq_len=total_seq_len,
-            encoder_hidden_states=None,
-            cached_encoder_key_value=cached_encoder_key_value,
-            attention_mask=None,
-        )
-        lm_logits = self.proj_out(hidden_states)
-        return lm_logits

From c88231ba031ceeb44b95bd264d83b1e6202819c3 Mon Sep 17 00:00:00 2001
From: Lesheng Jin <leshenj15@gmail.com>
Date: Fri, 1 Mar 2024 07:10:28 +0000
Subject: [PATCH 3/3] whisper checkpoint

---
 python/mlc_chat/__init__.py                   |    3 +-
 python/mlc_chat/base.py                       |    5 +-
 python/mlc_chat/compiler_pass/pipeline.py     |    1 +
 python/mlc_chat/interface/compile.py          |    1 +
 python/mlc_chat/model/model.py                |   14 +
 python/mlc_chat/model/whisper/__init__.py     |    0
 .../mlc_chat/model/whisper/whisper_loader.py  |   51 +
 .../mlc_chat/model/whisper/whisper_model.py   |  451 ++
 .../model/whisper/whisper_quantization.py     |   40 +
 .../quantization/group_quantization.py        |   41 +-
 test_cli.py                                   |   32 +
 test_disco.py                                 |   37 +
 test_disco_batched.py                         |   85 +
 test_gpt.sh                                   |   22 +
 test_llama.sh                                 |   20 +
 test_mistral.sh                               |   22 +
 test_multi_gpu.sh                             |   14 +
 test_old.sh                                   |    5 +
 test_phi.sh                                   |   30 +
 test_red_pajama.sh                            |   20 +
 test_slm.py                                   |   91 +
 test_slm_disco.py                             |  132 +
 test_whisper.py                               |  205 +
 test_whisper.sh                               |   21 +
 test_yi.sh                                    |   19 +
 tests/python/test_model_whisper.py            |   13 +-
 wasted/1.py                                   | 4652 +++++++++++++++++
 wasted/codetest1.py                           |   24 +
 wasted/deploy_mlc.py                          |  102 +
 wasted/deploy_new_mlc.py                      |  110 +
 wasted/disco_test/0.npy                       |  Bin 0 -> 35968 bytes
 wasted/disco_test/1.npy                       |  Bin 0 -> 35968 bytes
 wasted/disco_test/2.npy                       |  Bin 0 -> 35968 bytes
 wasted/disco_test/3.npy                       |  Bin 0 -> 35968 bytes
 wasted/disco_test/single.npy                  |  Bin 0 -> 35968 bytes
 wasted/gtable_to_json.py                      |   18 +
 wasted/load_np.py                             |   56 +
 wasted/model_info.json                        |    1 +
 wasted/phi_model_old.py                       |  451 ++
 wasted/sample_mlc_chat.py                     |   27 +
 wasted/sheet.csv                              |   29 +
 wasted/test000.py                             |   54 +
 wasted/test_7mistral.sh                       |   11 +
 wasted/test_awq.sh                            |   10 +
 wasted/test_awq_gemm.sh                       |   14 +
 wasted/test_decode_awq_gemm.py                |  110 +
 wasted/test_mistral.py                        |   54 +
 wasted/test_mistral_multi_gpu.sh              |   19 +
 wasted/test_mlc.py                            |   30 +
 wasted/test_mlc_hf.py                         |   23 +
 wasted/test_mlc_hf_batch.py                   |   81 +
 wasted/test_mlc_phi.py                        |   16 +
 wasted/test_mlc_red_pajama.py                 |   14 +
 wasted/test_old_red_pajama.sh                 |    3 +
 wasted/test_phi_hf.py                         |   19 +
 wasted/test_phi_mlc.py                        |   67 +
 wasted/test_preshard.sh                       |    7 +
 wasted/worker.py                              |   64 +
 58 files changed, 7416 insertions(+), 25 deletions(-)
 create mode 100644 python/mlc_chat/model/whisper/__init__.py
 create mode 100644 python/mlc_chat/model/whisper/whisper_loader.py
 create mode 100644 python/mlc_chat/model/whisper/whisper_model.py
 create mode 100644 python/mlc_chat/model/whisper/whisper_quantization.py
 create mode 100644 test_cli.py
 create mode 100644 test_disco.py
 create mode 100644 test_disco_batched.py
 create mode 100755 test_gpt.sh
 create mode 100755 test_llama.sh
 create mode 100755 test_mistral.sh
 create mode 100755 test_multi_gpu.sh
 create mode 100755 test_old.sh
 create mode 100755 test_phi.sh
 create mode 100755 test_red_pajama.sh
 create mode 100644 test_slm.py
 create mode 100644 test_slm_disco.py
 create mode 100644 test_whisper.py
 create mode 100755 test_whisper.sh
 create mode 100755 test_yi.sh
 create mode 100644 wasted/1.py
 create mode 100644 wasted/codetest1.py
 create mode 100644 wasted/deploy_mlc.py
 create mode 100644 wasted/deploy_new_mlc.py
 create mode 100644 wasted/disco_test/0.npy
 create mode 100644 wasted/disco_test/1.npy
 create mode 100644 wasted/disco_test/2.npy
 create mode 100644 wasted/disco_test/3.npy
 create mode 100644 wasted/disco_test/single.npy
 create mode 100644 wasted/gtable_to_json.py
 create mode 100644 wasted/load_np.py
 create mode 100644 wasted/model_info.json
 create mode 100644 wasted/phi_model_old.py
 create mode 100644 wasted/sample_mlc_chat.py
 create mode 100644 wasted/sheet.csv
 create mode 100644 wasted/test000.py
 create mode 100755 wasted/test_7mistral.sh
 create mode 100755 wasted/test_awq.sh
 create mode 100755 wasted/test_awq_gemm.sh
 create mode 100644 wasted/test_decode_awq_gemm.py
 create mode 100644 wasted/test_mistral.py
 create mode 100644 wasted/test_mistral_multi_gpu.sh
 create mode 100644 wasted/test_mlc.py
 create mode 100644 wasted/test_mlc_hf.py
 create mode 100644 wasted/test_mlc_hf_batch.py
 create mode 100644 wasted/test_mlc_phi.py
 create mode 100644 wasted/test_mlc_red_pajama.py
 create mode 100755 wasted/test_old_red_pajama.sh
 create mode 100644 wasted/test_phi_hf.py
 create mode 100644 wasted/test_phi_mlc.py
 create mode 100755 wasted/test_preshard.sh
 create mode 100644 wasted/worker.py

diff --git a/python/mlc_chat/__init__.py b/python/mlc_chat/__init__.py
index f577e0308e..351274a2f3 100644
--- a/python/mlc_chat/__init__.py
+++ b/python/mlc_chat/__init__.py
@@ -2,6 +2,7 @@
 
 MLC Chat is the app runtime of MLC LLM.
 """
-from . import protocol, serve
+
+# from . import protocol, serve
 from .chat_module import ChatConfig, ChatModule, ConvConfig, GenerationConfig
 from .libinfo import __version__
diff --git a/python/mlc_chat/base.py b/python/mlc_chat/base.py
index 13c7ba9f84..b5db400d79 100644
--- a/python/mlc_chat/base.py
+++ b/python/mlc_chat/base.py
@@ -1,4 +1,5 @@
 """Load MLC LLM library and _ffi_api functions."""
+
 import ctypes
 import os
 import sys
@@ -24,5 +25,5 @@ def _load_mlc_llm_lib():
 
 
 # only load once here
-if SKIP_LOADING_MLCLLM_SO == "0":
-    _LIB, _LIB_PATH = _load_mlc_llm_lib()
+# if SKIP_LOADING_MLCLLM_SO == "0":
+#     _LIB, _LIB_PATH = _load_mlc_llm_lib()
diff --git a/python/mlc_chat/compiler_pass/pipeline.py b/python/mlc_chat/compiler_pass/pipeline.py
index f7d0e0e8ab..fc8971bc55 100644
--- a/python/mlc_chat/compiler_pass/pipeline.py
+++ b/python/mlc_chat/compiler_pass/pipeline.py
@@ -1,4 +1,5 @@
 """The compilation pipeline for LLM applications."""
+
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 
diff --git a/python/mlc_chat/interface/compile.py b/python/mlc_chat/interface/compile.py
index 7f7d8a28d3..8a043f97e2 100644
--- a/python/mlc_chat/interface/compile.py
+++ b/python/mlc_chat/interface/compile.py
@@ -1,4 +1,5 @@
 """Python entrypoint of compilation."""
+
 import dataclasses
 import math
 from io import StringIO
diff --git a/python/mlc_chat/model/model.py b/python/mlc_chat/model/model.py
index a7fefdeca3..e6c3e4b41c 100644
--- a/python/mlc_chat/model/model.py
+++ b/python/mlc_chat/model/model.py
@@ -17,6 +17,7 @@
 from .phi import phi_loader, phi_model, phi_quantization
 from .qwen import qwen_loader, qwen_model, qwen_quantization
 from .stable_lm import stablelm_loader, stablelm_model, stablelm_quantization
+from .whisper import whisper_loader, whisper_model, whisper_quantization
 
 ModelConfig = Any
 """A ModelConfig is an object that represents a model architecture. It is required to have
@@ -195,4 +196,17 @@ class Model:
             "group-quant": stablelm_quantization.group_quant,
         },
     ),
+    "whisper": Model(
+        name="whisper",
+        model=whisper_model.WhisperForConditionalGeneration,
+        config=whisper_model.WhisperConfig,
+        source={
+            "huggingface-torch": whisper_loader.huggingface,
+            "huggingface-safetensor": whisper_loader.huggingface,
+        },
+        quantize={
+            "no-quant": whisper_quantization.no_quant,
+            "group-quant": whisper_quantization.group_quant,
+        },
+    ),
 }
diff --git a/python/mlc_chat/model/whisper/__init__.py b/python/mlc_chat/model/whisper/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/python/mlc_chat/model/whisper/whisper_loader.py b/python/mlc_chat/model/whisper/whisper_loader.py
new file mode 100644
index 0000000000..a5394d39ed
--- /dev/null
+++ b/python/mlc_chat/model/whisper/whisper_loader.py
@@ -0,0 +1,51 @@
+"""
+This file specifies how MLC's Whisper parameter maps from other formats, for example HuggingFace
+PyTorch, HuggingFace safetensors.
+"""
+
+import functools
+
+from mlc_chat.loader import ExternMapping
+from mlc_chat.quantization import Quantization
+
+from .whisper_model import WhisperConfig, WhisperForConditionalGeneration
+
+
+def huggingface(model_config: WhisperConfig, quantization: Quantization) -> ExternMapping:
+    """Returns a parameter mapping that maps from the names of MLC LLM parameters to
+    the names of HuggingFace PyTorch parameters.
+
+    Parameters
+    ----------
+    model_config : WhisperConfig
+        The configuration of the GPTNeoX model.
+
+    quantization : Quantization
+        The quantization configuration.
+
+    Returns
+    -------
+    param_map : ExternMapping
+        The parameter mapping from MLC to HuggingFace PyTorch.
+    """
+    model = WhisperForConditionalGeneration(model_config)
+    if quantization is not None:
+        model.to(quantization.model_dtype)
+    _, _named_params, _ = model.export_tvm(  # type: ignore[misc]
+        spec=model.get_default_spec(),
+        allow_extern=True,
+    )
+    named_parameters = dict(_named_params)
+
+    mapping = ExternMapping()
+
+    for mlc_name, mlc_param in named_parameters.items():
+        mapping.add_mapping(
+            mlc_name,
+            [mlc_name],
+            functools.partial(
+                lambda x, dtype: x.astype(dtype),
+                dtype=mlc_param.dtype,
+            ),
+        )
+    return mapping
diff --git a/python/mlc_chat/model/whisper/whisper_model.py b/python/mlc_chat/model/whisper/whisper_model.py
new file mode 100644
index 0000000000..431ee7fce7
--- /dev/null
+++ b/python/mlc_chat/model/whisper/whisper_model.py
@@ -0,0 +1,451 @@
+"""
+Implementation for Whisper architecture.
+"""
+
+import dataclasses
+from typing import Any, Dict, List, Optional, Tuple
+
+import tvm
+from tvm import te, tir
+from tvm.relax.frontend import nn
+
+from mlc_chat import op as op_ext
+from mlc_chat.support import logging
+from mlc_chat.support.config import ConfigBase
+
+logger = logging.getLogger(__name__)
+
+
+@dataclasses.dataclass
+class WhisperConfig(ConfigBase):
+    """Configuration of the Whipser model."""
+
+    vocab_size: int
+    num_hidden_layers: int
+    num_mel_bins: int
+    encoder_layers: int
+    encoder_attention_heads: int
+    decoder_layers: int
+    decoder_attention_heads: int
+    decoder_ffn_dim: int
+    encoder_ffn_dim: int
+    decoder_start_token_id: int
+    d_model: int
+    max_source_positions: int
+    max_target_positions: int
+    pad_token_id: int
+    bos_token_id: int
+    eos_token_id: int
+    begin_suppress_tokens: List[int]
+    suppress_tokens: List[int] = dataclasses.field(default_factory=list)
+    forced_decoder_ids: List[int] = dataclasses.field(default_factory=list)
+    tensor_parallel_shards: int = 1
+    kwargs: Dict[str, Any] = dataclasses.field(default_factory=dict)
+
+    def __post_init__(self):
+        self.context_window_size = self.max_source_positions
+        self.prefill_chunk_size = self.max_source_positions
+
+
+# pylint: disable=invalid-name,missing-docstring
+
+
+class WhisperPositionalEmbedding(nn.Module):
+    def __init__(self, max_seq_len: int, embed_dim: int):
+        super().__init__()
+        self.max_seq_len = max_seq_len
+        self.embed_dim = embed_dim
+        self.weight = nn.Parameter(
+            (max_seq_len, embed_dim),
+        )
+
+    def forward(self, x: nn.Tensor, offset: tir.Var):
+        def te_op(x: te.Tensor, embed: te.Tensor, offset: tir.Var):
+            def compute(_: tir.Var, j: tir.Var, k: tir.Var):
+                return embed[offset + j, k]
+
+            return tvm.te.compute([*x.shape, embed.shape[-1]], compute, name="position_embedding")
+
+        pos_embed = nn.tensor_expr_op(te_op, "position_embedding", args=[x, self.weight, offset])
+        return pos_embed
+
+
+class WhisperAttention(nn.Module):
+    def __init__(self, embed_dim: int, num_heads: int, kv_cache_len: int, bias: bool = True):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        self.head_dim = self.embed_dim // self.num_heads
+
+        if (self.head_dim * num_heads) != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
+                f" and `num_heads`: {num_heads})."
+            )
+        self.scaling = self.head_dim**-0.5
+        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=bias)
+        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=bias)
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=bias)
+        if kv_cache_len > 0:
+            self.k_cache = nn.KVCache(kv_cache_len, [self.num_heads, self.head_dim])
+            self.v_cache = nn.KVCache(kv_cache_len, [self.num_heads, self.head_dim])
+
+    def forward(
+        self,
+        hidden_states: nn.Tensor,
+        key_value_states: Optional[nn.Tensor] = None,
+        cached_cross_attn_states: Optional[Tuple[nn.Tensor, nn.Tensor]] = None,
+        attention_mask: Optional[nn.Tensor] = None,
+        total_seq_len: Optional[tir.Var] = None,
+    ) -> Tuple[nn.Tensor, Optional[Tuple[nn.Tensor, nn.Tensor]]]:
+        is_cross_attention = key_value_states is not None or cached_cross_attn_states is not None
+
+        h, d = self.num_heads, self.head_dim
+        bsz, q_len, _ = hidden_states.shape
+        assert bsz == 1, "Only support batch size 1 at this moment."
+
+        q = nn.reshape(self.q_proj(hidden_states) * self.scaling, (bsz, q_len, h, d))
+
+        if is_cross_attention:
+            # cross attention
+            if cached_cross_attn_states is None:
+                # no cache, cross attentions
+                kv_len = key_value_states.shape[1]
+                k = nn.reshape(self.k_proj(key_value_states), (bsz, kv_len, h, d))
+                v = nn.reshape(self.v_proj(key_value_states), (bsz, kv_len, h, d))
+                cached_kv = (k, v)
+            else:
+                # reuse cached k,v, cross_attentions
+                k, v = cached_cross_attn_states
+        else:
+            # self attention
+            k = nn.reshape(self.k_proj(hidden_states), (bsz, q_len, h, d))
+            v = nn.reshape(self.v_proj(hidden_states), (bsz, q_len, h, d))
+
+            if total_seq_len is not None:
+                # reuse cached k, v, self_attention
+                self.k_cache.append(nn.squeeze(k, axis=0))
+                self.v_cache.append(nn.squeeze(v, axis=0))
+                k = nn.reshape(self.k_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
+                v = nn.reshape(self.v_cache.view(total_seq_len), (bsz, total_seq_len, h, d))
+            else:
+                # encode self attention, no cache
+                ...
+
+        q = nn.permute_dims(q, [0, 2, 1, 3])  # [b, h, q_len, d]
+        k = nn.permute_dims(k, [0, 2, 1, 3])  # [b, h, q_len, d]
+        v = nn.permute_dims(v, [0, 2, 1, 3])  # [b, h, q_len, d]
+
+        attn_weights = nn.matmul(q, (nn.permute_dims(k, [0, 1, 3, 2])))  # [b, h, q_len, q_len]
+
+        if attention_mask is not None:
+            attn_weights = attn_weights + attention_mask
+
+        dtype = attn_weights.dtype
+        attn_weights = attn_weights.maximum(tir.min_value(dtype))
+        attn_weights = attn_weights.minimum(tir.max_value(dtype))
+        if dtype == "float32":
+            attn_weights = nn.softmax(attn_weights, axis=-1)
+        else:
+            attn_weights = nn.softmax(attn_weights.astype("float32"), axis=-1).astype(dtype)
+        attn_output = nn.matmul(attn_weights, v)  # [b, h, q_len, d]
+
+        attn_output = nn.permute_dims(attn_output, [0, 2, 1, 3])  # [b, q_len, h, d]
+        attn_output = nn.reshape(attn_output, (bsz, q_len, self.embed_dim))  # [b, q_len, h * d]
+
+        attn_output = self.out_proj(attn_output)
+
+        if is_cross_attention and cached_cross_attn_states is None:
+            return attn_output, cached_kv
+
+        return attn_output, None
+
+
+class EncoderLayer(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.self_attn = WhisperAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.encoder_attention_heads,
+            kv_cache_len=0,  # no need for kv_cache
+        )
+        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.fc1 = nn.Linear(self.embed_dim, config.encoder_ffn_dim)
+        self.fc2 = nn.Linear(config.encoder_ffn_dim, self.embed_dim)
+        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
+
+    def forward(
+        self, hidden_states: nn.Tensor, attention_mask: Optional[nn.Tensor] = None
+    ) -> nn.Tensor:
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            key_value_states=None,
+            cached_cross_attn_states=None,
+            attention_mask=attention_mask,
+            total_seq_len=None,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = nn.gelu(self.fc1(hidden_states))
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = residual + hidden_states
+        hidden_states = hidden_states.maximum(tir.min_value(hidden_states.dtype))
+        hidden_states = hidden_states.minimum(tir.max_value(hidden_states.dtype))
+
+        return hidden_states
+
+
+class DecoderLayer(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+        self.embed_dim = config.d_model
+
+        self.self_attn = WhisperAttention(
+            embed_dim=self.embed_dim,
+            num_heads=config.decoder_attention_heads,
+            kv_cache_len=config.max_target_positions,
+        )
+
+        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.encoder_attn = WhisperAttention(
+            self.embed_dim,
+            config.decoder_attention_heads,
+            kv_cache_len=config.max_target_positions,
+        )
+        self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.fc1 = nn.Linear(self.embed_dim, config.decoder_ffn_dim)
+        self.fc2 = nn.Linear(config.decoder_ffn_dim, self.embed_dim)
+        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
+
+    def forward(
+        self,
+        hidden_states: nn.Tensor,
+        encoder_hidden_states: nn.Tensor,
+        cached_encoder_hidden_states: nn.Tensor,
+        total_seq_len: tir.Var,
+        attention_mask: Optional[nn.Tensor] = None,
+        encoder_attention_mask: Optional[nn.Tensor] = None,
+    ) -> Tuple[nn.Tensor, Optional[Tuple[nn.Tensor]]]:
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+
+        # Self Attention
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            total_seq_len=total_seq_len,
+            key_value_states=None,
+            cached_cross_attn_states=None,
+            attention_mask=attention_mask,
+        )
+        hidden_states = residual + hidden_states
+
+        # Cross-Attention Block
+        residual = hidden_states
+        hidden_states = self.encoder_attn_layer_norm(hidden_states)
+        hidden_states, cross_attn_key_value = self.encoder_attn(
+            hidden_states=hidden_states,
+            total_seq_len=total_seq_len,
+            key_value_states=encoder_hidden_states,
+            cached_cross_attn_states=cached_encoder_hidden_states,
+            attention_mask=encoder_attention_mask,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = nn.gelu(self.fc1(hidden_states))
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = residual + hidden_states
+
+        if cached_encoder_hidden_states is None:
+            return hidden_states, cross_attn_key_value
+        return hidden_states, None
+
+
+class WhisperEncoder(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+
+        embed_dim = config.d_model
+        self.num_mel_bins = config.num_mel_bins
+        self.padding_idx = config.pad_token_id
+        self.max_source_positions = config.max_source_positions
+        self.embed_scale = 1.0
+
+        self.conv1 = nn.Conv1D(self.num_mel_bins, embed_dim, kernel_size=3, padding=1)
+        self.conv2 = nn.Conv1D(embed_dim, embed_dim, kernel_size=3, stride=2, padding=1)
+
+        self.embed_positions = nn.Embedding(self.max_source_positions, embed_dim)
+
+        self.layers = nn.ModuleList([EncoderLayer(config) for _ in range(config.encoder_layers)])
+        self.layer_norm = nn.LayerNorm(config.d_model)
+
+    def forward(self, input_features: nn.Tensor) -> nn.Tensor:
+        inputs_embeds = nn.gelu(self.conv1(input_features))
+        inputs_embeds = nn.gelu(self.conv2(inputs_embeds))
+
+        inputs_embeds = nn.permute_dims(inputs_embeds, [0, 2, 1])
+        embed_pos = self.embed_positions.weight
+
+        hidden_states = inputs_embeds + embed_pos
+
+        for idx, encoder_layer in enumerate(self.layers):
+            hidden_states = encoder_layer(hidden_states)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        return hidden_states
+
+
+class WhisperDecoder(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        super().__init__()
+
+        self.max_target_positions = config.max_target_positions
+        self.max_source_positions = config.max_source_positions
+        self.embed_scale = 1.0
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model)
+        self.embed_positions = WhisperPositionalEmbedding(self.max_target_positions, config.d_model)
+
+        self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(config.decoder_layers)])
+
+        self.layer_norm = nn.LayerNorm(config.d_model)
+
+    def forward(
+        self,
+        input_ids: nn.Tensor,
+        total_seq_len: Optional[tir.Var] = None,
+        encoder_hidden_states: Optional[nn.Tensor] = None,
+        cached_encoder_key_value: Optional[Tuple[Tuple[nn.Tensor]]] = None,
+        attention_mask: Optional[nn.Tensor] = None,
+    ):
+        # total_seq_len = Length of generated tokens
+        input_embeds = self.embed_tokens(input_ids)
+        past_seq_len = total_seq_len - 1
+        position_embeds = self.embed_positions(input_ids, offset=past_seq_len)
+
+        hidden_states = input_embeds + position_embeds
+
+        all_encoder_key_value = ()
+        for idx, decoder_layer in enumerate(self.layers):
+            ith_cached_encoder_key_value = (
+                cached_encoder_key_value[idx] if cached_encoder_key_value is not None else None
+            )
+            hidden_states, encoder_key_value = decoder_layer(
+                hidden_states=hidden_states,
+                total_seq_len=total_seq_len,
+                encoder_hidden_states=encoder_hidden_states,
+                cached_encoder_hidden_states=ith_cached_encoder_key_value,
+                attention_mask=attention_mask,
+            )
+            if cached_encoder_key_value is None:
+                all_encoder_key_value += (encoder_key_value,)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        if cached_encoder_key_value is None:
+            return hidden_states, all_encoder_key_value
+        else:
+            return hidden_states, None
+
+
+class WhisperModel(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        self.encoder = WhisperEncoder(config)
+        self.decoder = WhisperDecoder(config)
+
+
+class WhisperForConditionalGeneration(nn.Module):
+    def __init__(self, config: WhisperConfig):
+        self.config = config
+        self.model = WhisperModel(config)
+        self.proj_out = nn.Linear(config.d_model, config.vocab_size, bias=False)
+
+    def encode(self, input_features: nn.Tensor) -> nn.Tensor:
+        return self.model.encoder(input_features)
+
+    def decode(
+        self, input_ids: nn.Tensor, total_seq_len: int, encoder_hidden_states: nn.Tensor
+    ) -> Tuple[nn.Tensor, Tuple[Tuple[nn.Tensor]]]:
+        hidden_states, all_encoder_key_value = self.model.decoder.forward(
+            input_ids=input_ids,
+            total_seq_len=total_seq_len,
+            encoder_hidden_states=encoder_hidden_states,
+            cached_encoder_key_value=None,
+            attention_mask=None,
+        )
+        lm_logits = self.proj_out(hidden_states)
+        return lm_logits, all_encoder_key_value
+
+    def prefill(
+        self,
+        input_ids: nn.Tensor,
+        total_seq_len: int,
+        cached_encoder_key_value: Tuple[Tuple[nn.Tensor]],
+    ) -> nn.Tensor:
+        hidden_states, _ = self.model.decoder.forward(
+            input_ids=input_ids,
+            total_seq_len=total_seq_len,
+            encoder_hidden_states=None,
+            cached_encoder_key_value=cached_encoder_key_value,
+            attention_mask=None,
+        )
+        lm_logits = self.proj_out(hidden_states)
+        return lm_logits
+
+    def get_default_spec(self):
+        batch_size = 1
+        encode_input_ndim = 16000 * 30 // 160
+        mod_spec = {
+            "encode": {
+                "input_features": nn.spec.Tensor(
+                    [batch_size, self.config.num_mel_bins, encode_input_ndim], self.dtype
+                ),
+                "$": {
+                    "param_mode": "packed",
+                    "effect_mode": "packed",
+                },
+            },
+            "decode": {
+                "input_ids": nn.spec.Tensor([batch_size, "seq_len"], "int32"),
+                "total_seq_len": int,
+                "encoder_hidden_states": nn.spec.Tensor(
+                    [batch_size, self.config.max_source_positions, self.config.d_model], self.dtype
+                ),
+                "$": {
+                    "param_mode": "packed",
+                    "effect_mode": "packed",
+                },
+            },
+            "prefill": {
+                "input_ids": nn.spec.Tensor([batch_size, 1], "int32"),
+                "total_seq_len": int,
+                "cached_encoder_key_value": tuple(
+                    tuple(
+                        nn.spec.Tensor(
+                            [
+                                1,
+                                self.config.max_source_positions,
+                                self.config.decoder_attention_heads,
+                                self.config.d_model // self.config.decoder_attention_heads,
+                            ],
+                            self.dtype,
+                        )
+                        for i2 in range(2)
+                    )
+                    for i1 in range(self.config.num_hidden_layers)
+                ),
+                "$": {
+                    "param_mode": "packed",
+                    "effect_mode": "packed",
+                },
+            },
+        }
+        return nn.spec.ModuleSpec.from_raw(mod_spec, self)
diff --git a/python/mlc_chat/model/whisper/whisper_quantization.py b/python/mlc_chat/model/whisper/whisper_quantization.py
new file mode 100644
index 0000000000..1aafbe1082
--- /dev/null
+++ b/python/mlc_chat/model/whisper/whisper_quantization.py
@@ -0,0 +1,40 @@
+"""This file specifies how MLC's Whisper parameters are quantized using group quantization
+or other formats."""
+
+from typing import Tuple
+
+from tvm.relax.frontend import nn
+
+from mlc_chat.loader import QuantizeMapping
+from mlc_chat.quantization import GroupQuantize, NoQuantize
+
+from .whisper_model import WhisperConfig, WhisperForConditionalGeneration
+
+
+def group_quant(
+    model_config: WhisperConfig,
+    quantization: GroupQuantize,
+) -> Tuple[nn.Module, QuantizeMapping]:
+    """Quantize a GPTNeoX-architecture model using group quantization."""
+    model: nn.Module = WhisperForConditionalGeneration(model_config)
+    model.to(quantization.model_dtype)
+    model.dtype = quantization.model_dtype
+    quant_map = QuantizeMapping({}, {})
+    model = quantization.quantize_model(
+        model,
+        quant_map,
+        "",
+    )
+    return model, quant_map
+
+
+def no_quant(
+    model_config: WhisperConfig,
+    quantization: NoQuantize,
+) -> Tuple[nn.Module, QuantizeMapping]:
+    """Quantize a GPTNeoX model without quantization."""
+    model: nn.Module = WhisperForConditionalGeneration(model_config)
+    model.to(quantization.model_dtype)
+    model.dtype = quantization.model_dtype
+    quant_map = QuantizeMapping({}, {})
+    return model, quant_map
diff --git a/python/mlc_chat/quantization/group_quantization.py b/python/mlc_chat/quantization/group_quantization.py
index d854fd9cef..872aa49e9d 100644
--- a/python/mlc_chat/quantization/group_quantization.py
+++ b/python/mlc_chat/quantization/group_quantization.py
@@ -1,4 +1,5 @@
 """The group quantization config"""
+
 from dataclasses import dataclass
 from functools import partial
 from typing import Any, Callable, List, Literal, Optional, Tuple, Union
@@ -394,19 +395,25 @@ def forward(self, x: nn.Tensor) -> nn.Tensor:  # pylint: disable=invalid-name
                 weight,
                 scale,
                 axis=self.config.linear_quant_axis,
-                out_shape=[
-                    tir.IntImm("int64", self.out_features)
-                    if isinstance(self.out_features, int)
-                    else weight.shape[0],  # Reuse same tir.Var for symbolic shape (after Exporter)
-                    tir.IntImm("int64", self.in_features),
-                ]
-                if self.config.linear_weight_layout == "NK"
-                else [
-                    tir.IntImm("int64", self.in_features),
-                    tir.IntImm("int64", self.out_features)
-                    if isinstance(self.out_features, int)
-                    else weight.shape[1],  # Reuse same tir.Var for symbolic shape (after Exporter)
-                ],
+                out_shape=(
+                    [
+                        (
+                            tir.IntImm("int64", self.out_features)
+                            if isinstance(self.out_features, int)
+                            else weight.shape[0]
+                        ),  # Reuse same tir.Var for symbolic shape (after Exporter)
+                        tir.IntImm("int64", self.in_features),
+                    ]
+                    if self.config.linear_weight_layout == "NK"
+                    else [
+                        tir.IntImm("int64", self.in_features),
+                        (
+                            tir.IntImm("int64", self.out_features)
+                            if isinstance(self.out_features, int)
+                            else weight.shape[1]
+                        ),  # Reuse same tir.Var for symbolic shape (after Exporter)
+                    ]
+                ),
             ),
             name_hint="dequantize",
             args=[self.q_weight, self.q_scale],
@@ -485,9 +492,11 @@ def forward(self, x: nn.Tensor):  # pylint: disable=invalid-name
                 scale,
                 axis=-1,
                 out_shape=[
-                    tir.IntImm("int64", self.num)
-                    if isinstance(self.num, int)
-                    else weight.shape[0],  # Reuse same tir.Var for symbolic shape (after Exporter)
+                    (
+                        tir.IntImm("int64", self.num)
+                        if isinstance(self.num, int)
+                        else weight.shape[0]
+                    ),  # Reuse same tir.Var for symbolic shape (after Exporter)
                     tir.IntImm("int64", self.dim),
                 ],
             ),
diff --git a/test_cli.py b/test_cli.py
new file mode 100644
index 0000000000..f0313bddab
--- /dev/null
+++ b/test_cli.py
@@ -0,0 +1,32 @@
+help_str = """You can use the following special commands:
+  /help               print the special commands
+  /exit               quit the cli
+  /stats              print out the latest stats (token/sec)
+  /reset              restart a fresh chat
+  /reload [model]  reload model `model` from disk, or reload the current \
+model if `model` is not specified
+"""
+print(help_str)
+
+from prompt_toolkit import prompt
+from prompt_toolkit.key_binding import KeyBindings
+
+kb = KeyBindings()
+
+
+@kb.add("escape", "enter")
+def _(event):
+    event.current_buffer.insert_text("\n")
+
+
+@kb.add("enter")
+def _(event):
+    event.current_buffer.validate_and_handle()
+
+
+result = prompt(">", key_bindings=kb, multiline=True)
+print("12")
+print(result)
+
+# user_input = get_multiline_input()
+# print(user_input)
diff --git a/test_disco.py b/test_disco.py
new file mode 100644
index 0000000000..292fcc12c5
--- /dev/null
+++ b/test_disco.py
@@ -0,0 +1,37 @@
+# pylint: skip-file
+
+MODEL = "HF://junrushao/Llama-2-7b-chat-hf-q4f16_1-MLC"
+# MODEL = "HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC"
+# MODEL = "HF://junrushao/WizardCoder-15B-V1.0-q4f16_1-MLC"
+# MODEL = "HF://junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC"
+# MODEL = "HF://junrushao/phi-1_5-q4f16_1-MLC"
+# MODEL = "HF://junrushao/phi-2-q4f16_1-MLC"
+TP_SHARDS = 2
+import os
+
+from mlc_chat import ChatConfig, ChatModule, callback
+from mlc_chat.support import logging
+
+# temp_dir = "/opt/scratch/lesheng/mlc-llm/dist/tmp"
+
+# os.environ["MLC_TEMP_DIR"] = temp_dir
+# os.environ["MLC_CACHE_DIR"] = temp_dir
+# os.environ["MLC_JIT_POLICY"] = "REDO"
+
+
+logging.enable_logging()
+
+cm = ChatModule(
+    MODEL,
+    device="cuda",
+    chat_config=ChatConfig(
+        # context_window_size=1024,
+        # prefill_chunk_size=1024 if "Mistral" not in MODEL else 4096,
+        tensor_parallel_shards=TP_SHARDS,
+        # opt="flashinfer=0;cublas_gemm=1;cudagraph=0",
+    ),
+)
+cm.generate(
+    "What is the meaning of life?",
+    progress_callback=callback.StreamToStdout(callback_interval=2),
+)
diff --git a/test_disco_batched.py b/test_disco_batched.py
new file mode 100644
index 0000000000..0425e39788
--- /dev/null
+++ b/test_disco_batched.py
@@ -0,0 +1,85 @@
+# pylint: skip-file
+import json
+import logging
+import os
+import shutil
+
+import tvm
+
+from mlc_chat import ChatConfig, ChatModule
+
+temp_dir = "/opt/scratch/lesheng/mlc-llm/dist/disco_tmp"
+
+os.environ["MLC_TEMP_DIR"] = temp_dir
+os.environ["MLC_CACHE_DIR"] = temp_dir
+
+logging.basicConfig(
+    level=logging.INFO,
+    style="{",
+    datefmt="%Y-%m-%d %H:%M:%S",
+    format="[{asctime}] {levelname} {filename}:{lineno}: {message}",
+)
+
+
+def test_model(url, num_shards=1):
+    try:
+        cm = ChatModule(
+            url,
+            chat_config=ChatConfig(
+                context_window_size=1024,
+                max_gen_len=200,
+                tensor_parallel_shards=num_shards,
+            ),
+            device="cuda",
+        )
+        output = cm.generate(
+            # prompt="What is the meaning of life?",
+            prompt="Write a python program that computes a + b."
+        )
+    except:
+        output = "Failed"
+    return output
+
+
+print(test_model("HF://junrushao/gpt2-q4f16_1-MLC", 4))
+exit(0)
+
+with open("model_info.json", "r") as f:
+    model_dict = json.load(f)
+
+testout_dir = "disco_out"
+os.makedirs(testout_dir, exist_ok=True)
+
+# shutil.rmtree(temp_dir, ignore_errors=True)
+for model in model_dict:
+    model_infos = model_dict[model]
+    for quant in ["q4f16_1", "q4f32_1"]:  # , "q3f16_1"]:
+        url = model_infos[quant]
+        if len(url) == 0:
+            continue
+
+        print(f"{model}/{quant}: {url}")
+
+        os.makedirs(temp_dir, exist_ok=True)
+
+        model_out_dir = os.path.join(testout_dir, model)
+        os.makedirs(model_out_dir, exist_ok=True)
+        for i in [1, 2, 4]:  # , 8]:
+            print(i)
+            testout_file = os.path.join(model_out_dir, f"{quant}-{i}.out")
+
+            if os.path.exists(testout_file):
+                print(f"skip {testout_file}")
+                continue
+
+            output = test_model(url, i)
+            with open(testout_file, "w+") as f:
+                f.write(output + "\n")
+
+        shutil.rmtree(temp_dir, ignore_errors=True)
+
+        # if f"{model}_{quant}" in ["CodeLlama-7b-hf_q4f32_1"]:
+        #     continue
+        # if os.path.exists(testout_file):
+        #     print(f"skip {testout_file}")
+        #     continue
diff --git a/test_gpt.sh b/test_gpt.sh
new file mode 100755
index 0000000000..febd080f9b
--- /dev/null
+++ b/test_gpt.sh
@@ -0,0 +1,22 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+NUM_SHARDS=$1
+
+MODEL_PATH=/opt/scratch/lesheng/models/gpt2-medium
+OUTPUT_PATH=./dist/GPT-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template phi-2 -o $OUTPUT_PATH/params --tensor-parallel-shards $NUM_SHARDS
+
+python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --debug-dump $OUTPUT_PATH/debug
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+#"What is the meaning of life?"
+# CUDA_VISIBLE_DEVICES=4,5,6,7
+
+python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "a + b" --generate-length 256
diff --git a/test_llama.sh b/test_llama.sh
new file mode 100755
index 0000000000..1361365181
--- /dev/null
+++ b/test_llama.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+# QUANT=q4f16_0
+# QUANT=q4f16_ft
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf
+OUTPUT_PATH=./dist/new-llama-${QUANT}/
+NUM_SHARDS=2
+
+mkdir -p $OUTPUT_PATH
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template llama-2 -o $OUTPUT_PATH/params --tensor-parallel-shards $NUM_SHARDS
+
+# python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --opt "O2"
+
+# PRESHARD_NUM=$NUM_SHARDS python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256 --num-shards $NUM_SHARDS
diff --git a/test_mistral.sh b/test_mistral.sh
new file mode 100755
index 0000000000..972480ea6c
--- /dev/null
+++ b/test_mistral.sh
@@ -0,0 +1,22 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+MODEL_PATH=/opt/scratch/lesheng/models/Mistral-7B-Instruct-v0.2
+OUTPUT_PATH=./dist/mistral-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template mistral_default -o $OUTPUT_PATH/params
+
+# python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --opt "flashinfer=0;cublas_gemm=1;cudagraph=0" --overrides "context_window_size=32768;sliding_window_size=4096;prefill_chunk_size=4096;attention_sink_size=4"
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+# CUDA_VISIBLE_DEVICES=6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
+
+CUDA_VISIBLE_DEVICES=6,7 python -m mlc_chat.cli.benchmark \
+    --model /opt/scratch/lesheng/mlc-llm/dist/tmp/model_weights/junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC \
+    --model-lib /opt/scratch/lesheng/mlc-llm/dist/tmp/model_lib/a4e842ad2814e4e95c171a2685bdbab7.so \
+    --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/test_multi_gpu.sh b/test_multi_gpu.sh
new file mode 100755
index 0000000000..20d3b00347
--- /dev/null
+++ b/test_multi_gpu.sh
@@ -0,0 +1,14 @@
+#!/bin/bash
+set -e
+
+NUM_SHARDS=2
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+OUTPUT_PATH=./dist/new-llama-2-GPU/
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization q4f16_1 --conv-template llama-2 --context-window-size 4096 -o $OUTPUT_PATH --tensor-parallel-shards $NUM_SHARDS
+
+# python -m mlc_chat compile $OUTPUT_PATH -o $OUTPUT_PATH/llama.so
+
+python -m mlc_chat convert_weight $MODEL_PATH --quantization q4f16_1 -o $OUTPUT_PATH
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH --model-lib $OUTPUT_PATH/llama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/test_old.sh b/test_old.sh
new file mode 100755
index 0000000000..adcb99f87c
--- /dev/null
+++ b/test_old.sh
@@ -0,0 +1,5 @@
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python build.py --model $MODEL_PATH --quantization q4f16_1 --num-shards 2 --build-model-only --max-seq-len 4096 --use-cache 0
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python build.py --model $MODEL_PATH --quantization q4f16_1 --num-shards 2 --convert-weight-only --max-seq-len 4096 --use-cache 0
diff --git a/test_phi.sh b/test_phi.sh
new file mode 100755
index 0000000000..076d0381ba
--- /dev/null
+++ b/test_phi.sh
@@ -0,0 +1,30 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+NUM_SHARDS=$1
+
+MODEL_PATH=/opt/scratch/lesheng/phi-2
+OUTPUT_PATH=./dist/phi-2-${QUANT}-${NUM_SHARDS}/
+
+# MODEL_PATH=/opt/scratch/lesheng/models/phi-1_5
+# OUTPUT_PATH=./dist/phi-1_5-${QUANT}/
+
+# MODEL_PATH=/opt/scratch/lesheng/models/phi-1
+# OUTPUT_PATH=./dist/phi-1-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template phi-2 -o $OUTPUT_PATH/params --tensor-parallel-shards $NUM_SHARDS
+
+python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --debug-dump $OUTPUT_PATH/debug
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+#"What is the meaning of life?"
+# CUDA_VISIBLE_DEVICES=4,5,6,7
+
+python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "Print a + b" --generate-length 256
+
+# python -m mlc_chat.cli.benchmark --model ./dist/phi-2-q4f16_1-4/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "Print a + b" --generate-length 256
diff --git a/test_red_pajama.sh b/test_red_pajama.sh
new file mode 100755
index 0000000000..fbbaab4718
--- /dev/null
+++ b/test_red_pajama.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+MODEL_PATH=/opt/scratch/lesheng/models/RedPajama-INCITE-Chat-3B-v1
+OUTPUT_PATH=./dist/new-red-pajama-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template redpajama_chat -o $OUTPUT_PATH/params --tensor-parallel-shards 4
+
+# python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/red-pajama.so --debug-dump $OUTPUT_PATH/debug
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+# CUDA_VISIBLE_DEVICES=4,5,6,7
+python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/red-pajama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256 # --tensor-parallel-shards 2
+
+# python -m mlc_chat.cli.benchmark --model ./dist/new-red-pajama-q4f16_1/params --model-lib ./dist/new-red-pajama-q4f16_1/red-pajama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/test_slm.py b/test_slm.py
new file mode 100644
index 0000000000..35c8ca2ba4
--- /dev/null
+++ b/test_slm.py
@@ -0,0 +1,91 @@
+import argparse
+import os
+import pickle
+
+import numpy as np
+import torch
+import tvm
+from mlc_llm import utils
+from transformers import AutoTokenizer
+from tvm import relax
+
+
+def load_params(artifact_path: str, device, param_names):
+    from tvm.contrib import tvmjs  # pylint: disable=import-outside-toplevel
+
+    params, meta = tvmjs.load_ndarray_cache(artifact_path, device)
+    plist = []
+    size = meta["ParamSize"]
+    for pname in param_names:
+        plist.append(params[pname])
+    return plist
+
+
+class TestState:
+    def __init__(self, args):
+        self.primary_device = tvm.device(args.primary_device)
+        ex = tvm.runtime.load_module(args.model_lib)
+        self.vm = relax.VirtualMachine(ex, self.primary_device)
+        self.sess = None
+
+
+def deploy_to_pipeline(args) -> None:
+    state = TestState(args)
+    x = state.vm["_metadata"]()
+    metadata = eval(x)
+    params = metadata["params"]
+    param_names = []
+    for param in params:
+        param_names.append(param["name"])
+
+    primary_device = tvm.device(args.primary_device)
+    const_params = load_params(args.params, primary_device, param_names)
+    tokenizer = AutoTokenizer.from_pretrained(args.params, trust_remote_code=True)
+
+    print("Tokenizing...")
+    inputs = tokenizer(args.prompt, return_tensors="pt").input_ids.to(torch.int32).numpy()
+    print(inputs)
+    first_sampled_token = tvm.nd.array(np.array([[6234]]).astype("int32"), primary_device)
+    seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1]])
+    second_seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1] + 1])
+    kv_caches = state.vm["_initialize_effect"]()
+
+    print("Running inference...")
+    print("======================= Starts Encoding =======================")
+
+    prefill_func = state.vm["prefill"]
+
+    inputs = tvm.nd.array(inputs, device=primary_device)
+    logits, kv_caches = prefill_func(inputs, seq_len_shape, kv_caches, const_params)
+    print(logits)
+
+    exit(0)
+    print("======================= Starts Decoding =======================")
+    logits, kv_caches = state.vm["decode"](
+        first_sampled_token, second_seq_len_shape, kv_caches, const_params
+    )
+
+
+def _parse_args():
+    args = argparse.ArgumentParser()
+    args.add_argument("--model-lib", type=str, required=True)
+    args.add_argument("--params", type=str, required=True)
+    args.add_argument("--primary-device", type=str, default="auto")
+    args.add_argument("--prompt", type=str, default="What is the meaning of life?")
+    args.add_argument("--time-eval", default=False, action="store_true")
+    args.add_argument("--skip-rounds", type=int, default=0)
+    parsed = args.parse_args()
+
+    if parsed.primary_device == "auto":
+        if tvm.cuda().exist:
+            parsed.primary_device = "cuda"
+        elif tvm.metal().exist:
+            parsed.primary_device = "metal"
+        else:
+            raise ValueError("Cannot auto deduce device-name, please set it")
+    return parsed
+
+
+if __name__ == "__main__":
+    args = _parse_args()
+    deploy_to_pipeline(args)
diff --git a/test_slm_disco.py b/test_slm_disco.py
new file mode 100644
index 0000000000..849ae3ec2e
--- /dev/null
+++ b/test_slm_disco.py
@@ -0,0 +1,132 @@
+# pylint: skip-file
+import argparse
+import json
+import os
+
+import numpy as np
+import torch
+import tvm
+from transformers import AutoTokenizer
+from tvm import relax
+from tvm.relax.frontend.nn import IOEffect
+from tvm.runtime import disco as di
+
+import mlc_chat
+
+
+def load_params(artifact_path: str, device, param_names):
+    from tvm.contrib import tvmjs  # pylint: disable=import-outside-toplevel
+
+    params, meta = tvmjs.load_ndarray_cache(artifact_path, device)
+    plist = []
+    size = meta["ParamSize"]
+    for pname in param_names:
+        plist.append(params[pname])
+    return plist
+
+
+def CopyToWorker0(sess: di.ProcessSession, host_array):
+    array = sess.empty(host_array.shape, host_array.dtype)
+    sess.copy_to_worker_0(host_array, array)
+    return array
+
+
+def deploy_to_pipeline(args) -> None:
+    model_config = open(os.path.join(args.params, "mlc-chat-config.json")).read()
+
+    devices = list(range(args.tensor_parallel_shards))
+    sess = di.ProcessSession(
+        num_workers=args.tensor_parallel_shards, entrypoint="mlc_chat.cli.worker"
+    )
+    sess.init_ccl("nccl", *devices)
+    print(args.model_lib)
+    mod = sess.load_vm_module(args.model_lib)
+    create_kv_cache = mod["_initialize_effect"]
+    reset_kv = sess.get_global_func("vm.builtin.attention_kv_cache_array_clear")
+
+    prefill_func = mod["prefill"]
+
+    kv_caches = create_kv_cache()
+
+    # local_kv = kv_caches.debug_get_from_remote(0)
+    # print(type(local_kv))
+    # print(len(local_kv))
+    # exit(0)
+
+    loader = sess.get_global_func("mlc.loader.LoadMultiGPU")
+    params = loader(args.params, mod, model_config)
+
+    input_tokens_decode = sess.empty((1, 1), "int32")
+
+    tokenizer = AutoTokenizer.from_pretrained(args.params, trust_remote_code=True)
+    prompt_tokens = tokenizer(args.prompt, return_tensors="pt").input_ids.to(torch.int32).numpy()
+    print(prompt_tokens)
+    total_seq_len = prompt_tokens.shape[1]
+    prompt_tokens = tvm.nd.array(prompt_tokens)
+
+    sess.sync_worker_0()
+    # ForwardTokens
+    input_data = CopyToWorker0(sess, prompt_tokens)
+    seq_len_shape = tvm.runtime.ShapeTuple([total_seq_len])
+    ret = prefill_func(input_data, seq_len_shape, kv_caches, params)
+
+    logits_on_devices = ret.debug_get_from_remote(0)[0]
+    print(logits_on_devices)
+    total_seq_len += 1
+    exit(0)
+    second_seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1] + 1])
+
+    print("Running inference...")
+    print("======================= Starts Encoding =======================")
+
+    try:
+        prefill_func = state.vm["prefill"]
+    except AttributeError:
+        prefill_func = None
+
+    if inputs.shape[1] > 1 and prefill_func:
+        inputs = tvm.nd.array(inputs, device=primary_device)
+        logits, kv_caches = prefill_func(inputs, seq_len_shape, kv_caches, const_params)
+    else:
+        for i in range(inputs.shape[1]):
+            input_slice = tvm.nd.array(inputs[:, i : i + 1], device=primary_device)
+            logits, kv_caches = state.vm["decode"](
+                input_slice, seq_len_shape, kv_caches, const_params
+            )
+
+    exit(0)
+    print("======================= Starts Decoding =======================")
+    logits, kv_caches = state.vm["decode"](
+        first_sampled_token, second_seq_len_shape, kv_caches, const_params
+    )
+
+
+def _parse_args():
+    # @tvm.register_func("vm.builtin.debug_print")
+    # def _print(lineo: str, array) -> None:
+    #     print(f"{lineo}: shape = {array.shape}, dtype = {array.dtype}, data =\n{array}")
+
+    args = argparse.ArgumentParser()
+    args.add_argument("--model-lib", type=str, required=True)
+    args.add_argument("--params", type=str, required=True)
+    args.add_argument("--tensor-parallel-shards", type=int, default=4)
+    args.add_argument("--primary-device", type=str, default="auto")
+    args.add_argument("--prompt", type=str, default="What is the meaning of life?")
+    args.add_argument("--time-eval", default=False, action="store_true")
+    args.add_argument("--skip-rounds", type=int, default=0)
+    parsed = args.parse_args()
+
+    if parsed.primary_device == "auto":
+        if tvm.cuda().exist:
+            parsed.primary_device = "cuda"
+        elif tvm.metal().exist:
+            parsed.primary_device = "metal"
+        else:
+            raise ValueError("Cannot auto deduce device-name, please set it")
+    return parsed
+
+
+if __name__ == "__main__":
+    args = _parse_args()
+    deploy_to_pipeline(args)
+    deploy_to_pipeline(args)
diff --git a/test_whisper.py b/test_whisper.py
new file mode 100644
index 0000000000..0dc9593cf4
--- /dev/null
+++ b/test_whisper.py
@@ -0,0 +1,205 @@
+# pylint: disable=invalid-name,missing-docstring
+import json
+import time
+from typing import Any, Tuple
+
+import numpy as np
+import torch
+import tvm
+from datasets import load_dataset
+from scipy.io import wavfile
+from transformers import WhisperForConditionalGeneration as hf_Whisper
+from transformers import WhisperProcessor
+from tvm import relax
+
+from mlc_chat.model.whisper.whisper_model import (
+    WhisperConfig,
+    WhisperForConditionalGeneration,
+)
+
+
+def load_params(artifact_path: str, device, param_names):
+    from tvm.contrib import tvmjs  # pylint: disable=import-outside-toplevel
+
+    params, meta = tvmjs.load_ndarray_cache(artifact_path, device)
+    plist = []
+    size = meta["ParamSize"]
+    for pname in param_names:
+        plist.append(params[pname])
+    return plist
+
+
+def get_param_list(vm):
+    x = vm["_metadata"]()
+    metadata = eval(x)
+    params = metadata["params"]
+    param_names = []
+    for param in params:
+        param_names.append(param["name"])
+    return param_names
+
+
+def get_param_list000(config, dtype, submodel_name):
+    model = WhisperForConditionalGeneration(config)
+    # model.to(dtype)
+    encode_params = []
+    decode_params = []
+    prefill_params = []
+    for param, _ in model.named_parameters():
+        if param.startswith("model.encoder"):
+            encode_params.append(param)
+        else:
+            encode_params.append(param)
+            decode_params.append(param)
+            if "encoder_attn_k_proj" in param:
+                continue
+            if "encoder_attn_v_proj" in param:
+                continue
+            prefill_params.append(param)
+
+    return encode_params, decode_params, prefill_params
+
+
+def load_whisper_from_hf(model_name="openai/whisper-medium") -> Tuple[hf_Whisper, WhisperProcessor]:
+    processor = WhisperProcessor.from_pretrained(model_name)
+    # hf_model = hf_Whisper.from_pretrained(model_name)
+    # hf_model = hf_model.eval().to("cuda")
+    hf_model = None
+    return hf_model, processor
+
+
+def load_data(processor: WhisperProcessor, test_idx: int) -> Tuple[torch.Tensor, str]:
+    samplerate, data = wavfile.read("../librispeech_dummy.wav")
+    # print(samplerate)
+    # print(len(data))
+    input_features = processor(
+        data, sampling_rate=samplerate, return_tensors="pt"
+    ).input_features.to("cuda")
+    # print("input_features: ", input_features.shape)
+    return input_features, None
+
+
+def pipe(
+    model: Any, const_params: Any, config: WhisperConfig, input_features, device
+) -> torch.Tensor:
+    kv_caches = model["_initialize_effect"]()
+    # encode
+    encode_start = time.time()
+    encode_output, kv_caches = model["encode"](input_features, kv_caches, const_params)
+    device.sync()
+    encode_end = time.time()
+    print(f"encode {(encode_end - encode_start) * 1000:.2f}")
+
+    # decode start token
+    input_ids = torch.tensor([[config.decoder_start_token_id]], dtype=torch.int32)  # .to("cuda")
+    generated_tokens = [config.decoder_start_token_id]
+
+    input_ids = tvm.nd.array(input_ids, tvm.cuda())
+    while True:
+        decode_start = time.time()
+        if len(generated_tokens) == 1:
+            (outputs, encode_kv_cache), kv_caches = model["decode"](
+                input_ids,
+                tvm.runtime.ShapeTuple([len(generated_tokens)]),
+                encode_output,
+                kv_caches,
+                const_params,
+            )
+        else:
+            outputs, kv_caches = model["prefill"](
+                input_ids,
+                tvm.runtime.ShapeTuple([len(generated_tokens)]),
+                encode_kv_cache,
+                kv_caches,
+                const_params,
+            )
+        device.sync()
+        decode_end = time.time()
+        print(f"{(decode_end - decode_start) * 1000:.2f}")
+
+        outputs_logits = outputs.numpy()
+        next_token_logits = outputs_logits[:, 0, :]
+
+        # suppress tokens
+        next_tokens_scores = next_token_logits
+        next_tokens_scores[:, config.suppress_tokens] = -float("inf")
+
+        # suppress tokens at begin
+        if len(generated_tokens) == 1:  # + config.forced_decoder_ids[-1][0]:
+            next_tokens_scores[:, config.begin_suppress_tokens] = -float("inf")
+
+        # force tokens at sepcific position
+        generation_idx = len(generated_tokens)
+        current_token = dict(config.forced_decoder_ids).get(generation_idx, None)
+        if current_token is not None:
+            next_tokens_scores[:, :] = -float("inf")
+            next_tokens_scores[:, current_token] = 0
+
+        # argmax
+        next_token = np.argmax(next_tokens_scores, axis=-1)[0]
+        # next_token = torch.argmax(next_tokens_scores, dim=-1)[0]
+        # input_ids[0][0] = next_token
+        input_ids = tvm.nd.array(np.array([[next_token]], dtype=np.int32), tvm.cuda())
+
+        generated_tokens.append(next_token)
+
+        # stop when we meet eos_token_id or exceed the maximum length
+        if (
+            next_token == config.eos_token_id
+            or len(generated_tokens) == config.max_target_positions
+        ):
+            break
+
+    return generated_tokens
+
+
+def main():
+    with open("../models/whisper-large-v3/config.json", "r", encoding="utf-8") as config_file:
+        config = json.load(config_file)
+    config = WhisperConfig.from_dict(config)
+
+    model_dir = "dist/whisper-q0f16"
+    # Set the device and target
+    dev = tvm.cuda()
+    target = tvm.target.Target(
+        {
+            "kind": "cuda",
+            "max_shared_memory_per_block": dev.max_shared_memory_per_block,
+            "max_threads_per_block": dev.max_threads_per_block,
+            "thread_warp_size": dev.warp_size,
+            "registers_per_block": 65536,
+            "arch": "sm_" + tvm.cuda().compute_version.replace(".", ""),
+        }
+    )
+    ex = tvm.runtime.load_module(f"{model_dir}/model.so")
+    vm = relax.VirtualMachine(ex, dev)
+    param_list = get_param_list(vm)
+    const_params = load_params(f"{model_dir}/params", dev, param_list)
+
+    # load model from transformers
+    hf_model, processor = load_whisper_from_hf("../models/whisper-large-v3/")
+
+    # Test on librispeech_asr_dummy
+    input_features, text = load_data(processor, test_idx=0)
+    input_features = input_features.to("cpu").to(torch.float16)
+    input_features = tvm.nd.array(input_features, device=dev)
+
+    # const_params = [encode_params, decode_params, prefill_params]
+    generated_tokens = pipe(vm, const_params, config, input_features, dev)
+
+    # # compare with hf whisper output
+    # hf_predicted_ids = hf_model.generate(input_features).to("cpu")
+    # assert torch.equal(torch.tensor([generated_tokens], dtype=torch.long), hf_predicted_ids)
+
+    # decode token ids to text
+    output = processor.decode(generated_tokens, skip_special_tokens=True)
+    assert (
+        output
+        == " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel."
+    )
+
+    print("Transcription:\n", output)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/test_whisper.sh b/test_whisper.sh
new file mode 100755
index 0000000000..5698dd996e
--- /dev/null
+++ b/test_whisper.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+set -ex
+
+QUANT=q0f16
+# QUANT=q4f16_1
+# NUM_SHARDS=$1
+
+MODEL_PATH=/opt/scratch/lesheng/models/whisper-large-v3
+OUTPUT_PATH=./dist/whisper-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template phi-2 -o $OUTPUT_PATH/params
+
+python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --debug-dump $OUTPUT_PATH/debug --opt="cublas_gemm=0"
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+# CUDA_VISIBLE_DEVICES=4,5,6,7
+
+# python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "Print a + b" --generate-length 256
diff --git a/test_yi.sh b/test_yi.sh
new file mode 100755
index 0000000000..3fd7f35f3b
--- /dev/null
+++ b/test_yi.sh
@@ -0,0 +1,19 @@
+#!/bin/bash
+set -e
+
+# QUANT=q0f16
+QUANT=q4f16_1
+# QUANT=q4f16_0
+MODEL_PATH=../models/Yi-34B
+OUTPUT_PATH=./dist/Yi-34B-${QUANT}/
+NUM_SHARDS=1
+
+mkdir -p $OUTPUT_PATH
+
+python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template llama-2 -o $OUTPUT_PATH/params --tensor-parallel-shards $NUM_SHARDS
+
+python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/model.so --opt "flashinfer=0;cublas_gemm=0;cudagraph=0"
+
+python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/model.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256 --num-shards $NUM_SHARDS
diff --git a/tests/python/test_model_whisper.py b/tests/python/test_model_whisper.py
index 40515954a9..669543392b 100644
--- a/tests/python/test_model_whisper.py
+++ b/tests/python/test_model_whisper.py
@@ -1,16 +1,17 @@
 # pylint: disable=invalid-name,missing-docstring
-from typing import Tuple, Any
+from typing import Any, Tuple
 
 import torch
+import tvm
+from datasets import load_dataset
 from transformers import WhisperForConditionalGeneration as hf_Whisper
 from transformers import WhisperProcessor
-from datasets import load_dataset
-import torch
-
-import tvm
 from tvm.relax.frontend.nn import spec
 
-from mlc_llm.models.whisper import WhisperConfig, WhisperForConditionalGeneration
+from mlc_chat.model.whisper.whisper_model import (
+    WhisperConfig,
+    WhisperForConditionalGeneration,
+)
 
 
 def load_whisper_from_hf() -> Tuple[hf_Whisper, WhisperProcessor]:
diff --git a/wasted/1.py b/wasted/1.py
new file mode 100644
index 0000000000..0b50bebc41
--- /dev/null
+++ b/wasted/1.py
@@ -0,0 +1,4652 @@
+#  pylint: skip-file
+{
+    "model_type": "gpt_neox",
+    "quantization": "q4f16_1",
+    "context_window_size": 2048,
+    "prefill_chunk_size": 2048,
+    "sliding_window_size": -1,
+    "attention_sink_size": -1,
+    "tensor_parallel_shards": 2,
+    "params": [
+        {
+            "name": "gpt_neox.embed_in.q_weight",
+            "shape": [50432, 320],
+            "dtype": "uint32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.embed_in.q_scale",
+            "shape": [50432, 80],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.0.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.1.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.2.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.3.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.4.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.5.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.6.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.7.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.8.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.9.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.10.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.11.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.12.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.13.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.14.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.15.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.16.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.17.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.18.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.19.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.20.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.21.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.22.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.23.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.24.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.25.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.26.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.27.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.28.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.29.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.30.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.input_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.input_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.post_attention_layernorm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.post_attention_layernorm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.query_key_value.q_weight",
+            "shape": [3840, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_weight",
+                    "out_shape": [2, 3840, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.query_key_value.q_scale",
+            "shape": [3840, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_qkv_q_scale",
+                    "out_shape": [2, 3840, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.query_key_value.bias",
+            "shape": [3840],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.dense.q_weight",
+            "shape": [2560, 160],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_weight",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.dense.q_scale",
+            "shape": [2560, 40],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_q_scale",
+                    "out_shape": [2, 2560, 40],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.attention.dense.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_h_to_4h.q_weight",
+            "shape": [5120, 320],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_weight",
+                    "out_shape": [2, 5120, 320],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_h_to_4h.q_scale",
+            "shape": [5120, 80],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func": "_shard_dense_h_to_4h_q_scale",
+                    "out_shape": [2, 5120, 80],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_h_to_4h.bias",
+            "shape": [5120],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_4h_to_h.q_weight",
+            "shape": [2560, 640],
+            "dtype": "uint32",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_weight",
+                    "out_shape": [2, 2560, 640],
+                    "out_dtype": "uint32",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_4h_to_h.q_scale",
+            "shape": [2560, 160],
+            "dtype": "float16",
+            "preprocs": [
+                {
+                    "func_name": "_shard_dense_4h_to_h_q_scale",
+                    "out_shape": [2, 2560, 160],
+                    "out_dtype": "float16",
+                }
+            ],
+        },
+        {
+            "name": "gpt_neox.layers.31.mlp.dense_4h_to_h.bias",
+            "shape": [2560],
+            "dtype": "float32",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.final_layer_norm.weight",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {
+            "name": "gpt_neox.final_layer_norm.bias",
+            "shape": [2560],
+            "dtype": "float16",
+            "preprocs": [],
+        },
+        {"name": "embed_out.q_weight", "shape": [50432, 320], "dtype": "uint32", "preprocs": []},
+        {"name": "embed_out.q_scale", "shape": [50432, 80], "dtype": "float16", "preprocs": []},
+    ],
+    "memory_usage": {
+        "_initialize_effect": 0,
+        "decode": 16155140,
+        "prefill": 460539904,
+        "softmax_with_temperature": 0,
+    },
+}
diff --git a/wasted/codetest1.py b/wasted/codetest1.py
new file mode 100644
index 0000000000..6fcec18eba
--- /dev/null
+++ b/wasted/codetest1.py
@@ -0,0 +1,24 @@
+# pylint: skip-file
+cnt = 0
+
+
+def bt(nums, target, cur_i, cur_v):
+    global cnt
+    if cur_i == len(nums):
+        if cur_v == target:
+            cnt = cnt + 1
+        else:
+            pass
+    else:
+        bt(nums, target, cur_i + 1, cur_v + nums[cur_i])
+        bt(nums, target, cur_i + 1, cur_v - nums[cur_i])
+
+
+def f(nums, target):
+    bt(nums, target, cur_i=0, cur_v=0)
+    return cnt
+
+
+nums = [1, 1, 1, 1, 1]
+target = 3
+print(f(nums, target))
diff --git a/wasted/deploy_mlc.py b/wasted/deploy_mlc.py
new file mode 100644
index 0000000000..3a3ac9fb36
--- /dev/null
+++ b/wasted/deploy_mlc.py
@@ -0,0 +1,102 @@
+import argparse
+import os
+import pickle
+
+import numpy as np
+import torch
+import tvm
+from mlc_llm import utils
+from transformers import AutoTokenizer
+from tvm import relax
+
+
+class TestState:
+    def __init__(self, args):
+        self.primary_device = tvm.device(args.primary_device)
+        ex = tvm.runtime.load_module(
+            os.path.join(
+                args.artifact_path,
+                f"{args.model}-{args.quantization.name}-{args.primary_device}.so",
+            )
+        )
+        self.vm = relax.VirtualMachine(ex, self.primary_device)
+        self.sess = None
+
+
+def deploy_to_pipeline(args) -> None:
+    @tvm.register_func("debug_print")
+    def debug_print(dummy_object, info, a):
+        print(a.numpy())
+        return dummy_object
+
+    primary_device = tvm.device(args.primary_device)
+    const_params = utils.load_params(args.artifact_path, primary_device)
+    state = TestState(args)
+    tokenizer = AutoTokenizer.from_pretrained(
+        os.path.join(args.artifact_path, "params"), trust_remote_code=True
+    )
+
+    print("Tokenizing...")
+    inputs = tokenizer(args.prompt, return_tensors="pt").input_ids.to(torch.int32).numpy()
+    first_sampled_token = tvm.nd.array(np.array([[6234]]).astype("int32"), primary_device)
+    seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1]])
+    second_seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1] + 1])
+    dummy_obj = tvm.nd.array(np.array([[6234]]).astype("int32"), primary_device)
+    kv_caches = state.vm["create_kv_cache"]()
+
+    print("Running inference...")
+    print("======================= Starts Encoding =======================")
+
+    try:
+        prefill_func = state.vm["prefill"]
+    except AttributeError:
+        prefill_func = None
+
+    if inputs.shape[1] > 1 and prefill_func:
+        inputs = tvm.nd.array(inputs, device=primary_device)
+        logits, kv_caches, dummy_obj = prefill_func(
+            inputs, seq_len_shape, kv_caches, dummy_obj, const_params
+        )
+    else:
+        for i in range(inputs.shape[1]):
+            input_slice = tvm.nd.array(inputs[:, i : i + 1], device=primary_device)
+            logits, kv_caches = state.vm["decode"](
+                input_slice, seq_len_shape, kv_caches, const_params
+            )
+
+    exit(0)
+    print("======================= Starts Decoding =======================")
+    logits, kv_caches, dummy_obj = state.vm["decode"](
+        first_sampled_token, second_seq_len_shape, kv_caches, dummy_obj, const_params
+    )
+
+
+def _parse_args():
+    args = argparse.ArgumentParser()
+    args.add_argument("--local-id", type=str, required=True)
+    args.add_argument("--artifact-path", type=str, default="dist")
+    args.add_argument("--primary-device", type=str, default="auto")
+    args.add_argument("--prompt", type=str, default="What is the meaning of life?")
+    args.add_argument("--time-eval", default=False, action="store_true")
+    args.add_argument("--skip-rounds", type=int, default=0)
+    parsed = args.parse_args()
+    parsed.model, parsed.quantization = parsed.local_id.rsplit("-", 1)
+    utils.argparse_postproc_common(parsed)
+
+    parsed.artifact_path = os.path.join(
+        parsed.artifact_path, f"{parsed.model}-{parsed.quantization.name}"
+    )
+
+    if parsed.primary_device == "auto":
+        if tvm.cuda().exist:
+            parsed.primary_device = "cuda"
+        elif tvm.metal().exist:
+            parsed.primary_device = "metal"
+        else:
+            raise ValueError("Cannot auto deduce device-name, please set it")
+    return parsed
+
+
+if __name__ == "__main__":
+    args = _parse_args()
+    deploy_to_pipeline(args)
diff --git a/wasted/deploy_new_mlc.py b/wasted/deploy_new_mlc.py
new file mode 100644
index 0000000000..adf101ca7e
--- /dev/null
+++ b/wasted/deploy_new_mlc.py
@@ -0,0 +1,110 @@
+import argparse
+import os
+import pickle
+
+import numpy as np
+import torch
+import tvm
+from mlc_llm import utils
+from transformers import AutoTokenizer
+from tvm import relax
+
+
+def load_params(artifact_path: str, device, param_names):
+    from tvm.contrib import tvmjs  # pylint: disable=import-outside-toplevel
+
+    params, meta = tvmjs.load_ndarray_cache(f"{artifact_path}/params", device)
+    plist = []
+    size = meta["ParamSize"]
+    for pname in param_names:
+        plist.append(params[pname])
+    return plist
+
+
+class TestState:
+    def __init__(self, args):
+        self.primary_device = tvm.device(args.primary_device)
+        ex = tvm.runtime.load_module(os.path.join(args.artifact_path, f"model.so"))
+        self.vm = relax.VirtualMachine(ex, self.primary_device)
+        self.sess = None
+
+
+def deploy_to_pipeline(args) -> None:
+    state = TestState(args)
+    x = state.vm["_metadata"]()
+    metadata = eval(x)
+    params = metadata["params"]
+    param_names = []
+    for param in params:
+        param_names.append(param["name"])
+
+    primary_device = tvm.device(args.primary_device)
+    const_params = load_params(args.artifact_path, primary_device, param_names)
+    tokenizer = AutoTokenizer.from_pretrained(
+        os.path.join(args.artifact_path, "params"), trust_remote_code=True
+    )
+
+    print("Tokenizing...")
+    inputs = tokenizer(args.prompt, return_tensors="pt").input_ids.to(torch.int32).numpy()
+    print(inputs)
+    first_sampled_token = tvm.nd.array(np.array([[6234]]).astype("int32"), primary_device)
+    seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1]])
+    second_seq_len_shape = tvm.runtime.ShapeTuple([inputs.shape[1] + 1])
+    kv_caches = state.vm["_initialize_effect"]()
+
+    print("Running inference...")
+    print("======================= Starts Encoding =======================")
+
+    try:
+        prefill_func = state.vm["prefill"]
+    except AttributeError:
+        prefill_func = None
+
+    if inputs.shape[1] > 1 and prefill_func:
+        inputs = tvm.nd.array(inputs, device=primary_device)
+        logits, kv_caches = prefill_func(inputs, seq_len_shape, kv_caches, const_params)
+    else:
+        for i in range(inputs.shape[1]):
+            input_slice = tvm.nd.array(inputs[:, i : i + 1], device=primary_device)
+            logits, kv_caches = state.vm["decode"](
+                input_slice, seq_len_shape, kv_caches, const_params
+            )
+
+    exit(0)
+    print("======================= Starts Decoding =======================")
+    logits, kv_caches = state.vm["decode"](
+        first_sampled_token, second_seq_len_shape, kv_caches, const_params
+    )
+
+
+def _parse_args():
+    args = argparse.ArgumentParser()
+    args.add_argument("--local-id", type=str, default="")
+    args.add_argument("--artifact-path", type=str, default="dist")
+    args.add_argument("--primary-device", type=str, default="auto")
+    args.add_argument("--prompt", type=str, default="What is the meaning of life?")
+    args.add_argument("--time-eval", default=False, action="store_true")
+    args.add_argument("--skip-rounds", type=int, default=0)
+    parsed = args.parse_args()
+    # parsed.model, parsed.quantization = parsed.local_id.rsplit("-", 1)
+    parsed.model = "Phi-2"
+    parsed.quantization = "q0f16"
+    utils.argparse_postproc_common(parsed)
+
+    # parsed.artifact_path = os.path.join(
+    #     parsed.artifact_path, f"{parsed.model}-{parsed.quantization.name}"
+    # )
+
+    if parsed.primary_device == "auto":
+        if tvm.cuda().exist:
+            parsed.primary_device = "cuda"
+        elif tvm.metal().exist:
+            parsed.primary_device = "metal"
+        else:
+            raise ValueError("Cannot auto deduce device-name, please set it")
+    return parsed
+
+
+if __name__ == "__main__":
+    args = _parse_args()
+    deploy_to_pipeline(args)
diff --git a/wasted/disco_test/0.npy b/wasted/disco_test/0.npy
new file mode 100644
index 0000000000000000000000000000000000000000..a53de3487455b8ce9ef702678c5ad7dfa228dfad
GIT binary patch
literal 35968
zcmbSS<$K%M(=8-z3Q3bRV7Az@En|{0Gc$L~%(P`@ZquM0lO#*FY)iK6q|6N4Zeh#J
z%*_4e{TE(+)BSKCX`Z>~o^xg<zf14V-TGIoFss6R&B#$RMoiZ<05pw8TQvlr89inC
z-_wUp9x`S6$Wi~d-e%au8KcVAXN(;-byWE~ZXtlW1YpgnW7ZPDLg4@LRBsr)6Y*;|
zMtPzEA<ZVygPb{d67LcC7+gvmE4gJHmZecQG-qhnYbu%h^Oeng!|ior;CP+gkm&#6
zv6(4+FMkoguT&bJq01~mWh(raxerkwa{RB5U`{gMW}33{*o!D&NC^`R6*)(=v!Gl>
z;uilR`q2C#IDt<y-51TF+M$I+XY@94G0;qN*W5x-<tQVZA;TbAXmdw!ukLT=wG}mV
zDeV(BVUx|R*hl8EhGT@(_*Za`&{0>%P}MlXJdQ}Q9FLAPZ};_OKZTSP_ZWT{4n<zq
z{SfMDs%Cm)?Zq_`Dq$-^R_!axh~PY-9^Qg#le1iui-cjSmPTLH*|oJMx`ViFT_m&%
zugpH6CSqM;myJni3O=li@$@II3No$F+{3adTo60M_4QBG^TFf6gls?2MDtiIvA|He
z^`uA;6P-n5Qg}(WUYv|HjIGv0+0#NrZZmf-<S<vRCpSK`dJV$`Md*olwfSx23ev{1
z*Mvv6;zjVT0A;K~qy_)w&*gLzXA}9+emrR~^Nq|`!wl2JoEvIn*hTQ%+zeD##oEi-
zx@@hvAGh9m6Rko{&v<HHVM-JF=nnx0mDl|Bp<UcqO;v-$V!|gIt$1}|KIb+o`CPVM
zS`vREGC3L-ONI6k<47CXnz#@v)}#>UGee|W*o6Dd4`l>fGx|mx63RCf_|_VC5N6e+
z*oL~pun6JSoHLqYi}Z#-L!lktJ6vG?#jY+}z;_c?z*VS@u@-{bcZs8r#e5}=Otnqi
z!7UfYVC$k4t!!W#R$n}A{=je29p&aRlMOw@eWpq})ZEuH+0xx|K^&Ux_1`gV;N0L5
z%R7U=bgX5cImgh~(wk5Tv(jd852L%pcCMk?J3^s(YS|C+^wvk>`q)PQx8M!ya8r($
zW|_nn@_J*I)u7?giW-CngzH<Tgzxc3Ez|U!^sUXN!sT3*SZATkykA$JzkrsSep~wc
zCh{YVJz~dvM@x0FA^81RzHzzsDy%U7q$*Kwv$~Y-iCtq8^vS$hle6J;bP2IC^2>N2
zm`VROyyF4!v7pcu8&7Mk%uVY`?rBu-&NVmXCh=Lun&JiblK>L{#G8gImOWw{Z$Q+C
zFN*82zhd1?i5!tN&ag0gmr|&^DJ+H_{3vvqxzKzG&eB=4UJ7g^C%dLNmVx;x(KbH2
zZ)<oNw|GaIxWtrdKBv?1=lSL#W!dxqM$FK^B~mR*!sk4r!`s9Sk%r>N;4gdw*M7&y
zQ19Twu$!nNb~I;&+wd6{T|rCt9{i=HN!i`3+SEWxcLVN!8nQ-v3DziQ^qC7ycOxC~
zoKV+rZ$ssfO$<eMSW^ueZn$+@_C;izG0QwvtjG>De!vg=n|WtOrtl&^FZ$h3H?%g~
zO}`vS)^!M=;ws{50JBC7@z!nNz1Ves0iq<JTE91awv6|82D<E=hacU#CA3c6(l}Xr
z(EQd|mB}*!!Ue1wn$0~YehRDkt>$_D{@VM&?WXqH!qNlqX#R=yYSbAWVcr{jNK`Lf
zt$$%QnM0<0enGgqphjfQPDB@DQg&9LD!DMy!ThvT#!39sj4hI0YUdG^Jr#muO5HdZ
zN)6|T1iwU@r8{q^YpoD^Yc7sPIgPG)s3yHp;E3^jF#0+4-1Y-f;bkID-ZKucHV*0d
zZ}3fBC+>*l651mCH=fR%4+OoWGu<SHUlTcO``~xN73pNIK`daxH8*GNFjlv`=d<-n
z-@({z{7KYh{-FJa0ii3z=U_WQWBBe{fK`o77N>@e@oV`j5w&@&ehSi)uxx!sG%$V)
z_9prfPsG8&oN}ZKfPAE1R<R~SyvtV+J7mw|8U>v=9K9fI%nZwVgKvp8Q=cI!myXHm
zX|7=Ph|l?^o)0<kk(XSzh`8->`Io%LjtVRMjf1G7vw1o>D_{q{X-k1*Jej=`b*TQ0
zr0}RQ#dt#O363qB6gnIYijl}M)2i$_mT3k#|3ql2&(vY2`o?W+22oGE7_JeL>c5wL
z(M*X9jhMro#fuz+m?CQ;{l!tjtI#S`X?!IbgBQ6Qrbqe@v2i%=96`LLPcX-$TaEvL
zy@|QzEWb~vpgJG$Soc^M^AF3F=-4c4c#vhB_&RH-@wCv~)+cf#)F#x$SY7+vSS>V&
z|4)o*GL8R+(#!|KxDm|00)oNSdaJnAbV|E{e_tvXUs~1>*DYPGH1}5MUi8FuQ2Zin
zE&CDfMiiRuOs=_3_$&}_Ic~fbT`p7;uSR~d0M)dNHg|whwfnJQy2M}=-J!DU#&3qf
z;zevlc$sibmtq+yz6ajw=cy}$1d(FY5$EwD@RE4kn3*Dpv@!M!zY!V{iSV@S#}?R}
zn6;l95O`#MU8)WJ3LMab+(u%rVV&hu_SAqZI7=7Hs$}VJn7sXn(B1qY4EaZx3StB4
z8`z~#13}IFGVG<MZQT^e&HSOanlAB;J(J9t)=uHXz{k)!@d=^T7KOH#RxzzMp;!^o
zK3I>{1~=8+Q>KBHx?wRr)GqW)^S-oNtfzJ;zTY}nKcp;ZjAt5~`{HZN2K_<vRqbId
zEzH(^LuwmJxGAAKbbCW?7AS9szYjNvelaySSxQ^_QEo5K3Pr+jBQG|LW<)k?CE=~s
zzr-EU@qz8;<ECFiLvpaV)$k_r(v%sUAM9h!GQ8IQXSyXIB4Td3GePJhFyS#me*!jD
z3pMetGr{I^9Q2Oy<yzvRtC{(l$Jm;P!noU`Cf?C?#9vTPbEa=p^r`*{6XuMj2KXlh
zqaSOM@_D*>p$JxBcr2WX9YpGc)gaBSMrRw_S@-kmaCbdw8K|#k9UZ304YAEWOiLP&
zt+04Cu#Eo}iBpd>KFihxR|uQUdB#fC?Pwo;u5o2-C;o(~!l&xi8h;utnA(@V6o#1}
zT7zs`*lmn6=q>w%b$nM-34Z{&Zeg=(Fc0_#TwLTC-vc>ZBoC~Jnh>jy7X1_M9PMv#
zmQnP9oR{Ejb1ggxs~hEY)2x3BpFMB<cZeOPx8c3u;?iyA5}~vCmDOliWatswYpGvW
z>^_Q{g9LI9Ya81s_9SM+PH7WMcbY;NWVADX<7b8Y;Y!5V=r!CQoJ4$Ku8P<BgfeHO
zkNKpvYuScyVeFhRORO1A7boJ&jd8lInu@Vc!JTAkgo=I-Y0On}l0D6$+pr&c1NBMO
zS>I5<u5`EQbtK*~gq|ur%jfe8OB>_o2$@9#tu(LTL!saL9a4GeW78l$7F``3C`vOc
z#7O*DC`-``-)9(FtEIV{mNb{KCg!tP%W^TYKXO@EPrN1?I9mpaO(#vu!vlPGEC~j=
zRX}#JL8g`Er}nd+)9cNqSe&6(c%S+|^dNtAD`%F8l_LAiRrNV+pU}$CGPa_*h50d~
zw)A5i{4Gts>1bd!V^RFTs+o>D4uv`k*9=o4g0`AIkCXZ1m^tCK=6b=Oz6PPbXvGk3
zcpfaYHWP;N9m+7NQurj&(V%X7G-zlYs%1`&9Yh!5#X@iVlC`N&**}q7tCbnQg-;l2
zMDsb6kn$D7T}&H;HL~NQ{{-Vh+UQ+Pckhg>XR+Uqi7z#bHeL#x!1b0?>s3CP=z!i6
z7m3>~$GwC3-9C!AVt@s~dO!o=o9(|WF9l=tGkss1Ztci}{$=5~&_}oh(S%%Sm<7Ev
z{;laxJTNWf8(J=KA0?<ckxvP>EFEE<VO%IY1OtW>fjr=%K0<6ZbVY_+dRW($)%MN8
zS_(U2-^^Nk51=*I$l^;s^A_Qn=_WnU%*U>obIkW8|Cq;=^<%8ry@je-BgOedTk^25
zEZU1O=m>*bbn=%4#nxGY?1*28!<UC@X3d4`2k+}<_!FZF9cXICB>Mi-S2T9W?xX)=
zX&-eNI9;*jl#$PRA8soKwP8aGJ_#RcX2X-vCj=l)5AKb2LUXJSc%`|gE=ky7*csSh
zegkb4v>^w+$T&W0lJTr%V{|58y`-YPg4jSt8<pWPrSqao%#}ia@jtlN;s>q3PusSO
zev5A=8fv$RpA3^UUHA=X$efz>FtE+s)RbOo&x&QW!#IBrVFUX~yFWC+#75e|x5Ihj
z5Y<g_2KavGXLB9H&d3R47~_3w3_o;RgCos@3{Gv@w)dQa@2Gtu=0*oQ^+NaT^}(O`
z&_J;nwv5GEh7R*4)$D8>V#O1sF>$%Rwd-7<CT-vo@&5XGWLll4wj-7Wnl2F=pTj)i
zB!tu7$bj%Ov>U?Rv--sbY(q>Nf-8yYrcwAWasn=d<E@FbTBj#3=BSC05g^trEb#qB
z)G!QbBW|`_;mO!s%|qcNFEI;TFHtZ3%5h~iBN<^yxFS3j{tilwGSOs@H)F!NSYv#T
zZJQVnbvh78q>qdHG#!!QVr4#8G2fb<eN`jykA*V^58iF}U8|PO2v<uiWA@mt^L}q5
zlc|*A7aK3*<HTP?dvu}jg<llzW!{494yRa;nj2f5^8zu*+K(8bzg^mR2Pme~S7MLE
zPel9d&8bTB;K(!IZ}JTH!03pbGmH!Ix=uce?y>BUuU%|}>2_3MuBA=JZt9zcx@cib
zyV!Wcqk!Dn&oI!qH1-c#9LtQ%HB<=QH^)c13JKA9VZr(b-C>r}_jo*P7Fy~3W;gde
z<PkfR^fj(Pv+!T+A<G@}bgq%_9`{OHArNm~E6f+s$SD21$aGV;@W{yha696jt5(^4
z05|Q{K4)a58;p@{GXsxfuS4&(H_Y`S2~4}d4q{hXGwoBnx^akadsJokh1T{gA@-re
zk)_-ePG_!Tn(xP=e~S%#(}n$^Y}KmhLhU?L2kl3pnnh-~t6ve>?)ca8v2?x>5`G5z
zTgKyigx|5ZfmOKCd{Mu^J5Altm66pn^qh}qI|&8u;mC1TPN$m|>KbT|;c+U=28t=#
zmzIgfqlUNG>?m(Vte1tkzKi-M(dmY+Cfu|wq|8nb9b&Gbfn0;H&<XG~OR0GsQ-W2p
zt<n?xve*w1wI382;z-OGn9nAA7t*QpVnZEclmVke*5P1RPJ-IQABe`mI;BHl7Z4}*
z7Drkhvu&)Sh>77=)DG?h-$QH@-p&;nUK9V)wQzwsi_a1shI0J12v}&QzKH>RGcgrg
zj$Y$+$^e=e%OyVghY0<`vqA}G9UB)4MAM0N=6FqlrI&U9z?gRtH+=(y0p+_&`6|BQ
z?v6mr_X=x>71UT8^%H|kS-Ml2PT@>Vymbz-NcGaR*6=}77G0wcvwcmsOp}PpR@ht^
z>}>2U^Z`uii<G^21NUF(oj9dzm{v~R4b9NV6q)jkSX1*q!n5!ne3)USH#KUrR4Qtq
zonm@H3_#xoi^QMde}kZDPV~9h$<WGdP_=@p@^ZEy_E7nXe-=t+i%eO;wnbUiCc<=L
zI@ckLg_?#t8oC(nX&VqH!q3&XkSc6P+TwQW*>FeQFk-&-i(x;ZGwX!j+8@42;ww7U
zR9_6{rA9{*d$4Clok)n|(Y}$_rg(Qh1A>(WpRzacL}9R@AD==2>>zxRAt<~z&lJ`X
zqYRrOyTbE~^6(Fh1uMX_sTtZukdGK=-eZJI7x{YYp70V&2O}fx8@y%EfK#G&Lzqt!
z7@v`CE$-A8XtH8ChCJN>)r+D;%QqpDAESvAIy&YBS88R#V17~{6-Ds>h>lu^VuioI
zv8|z-H$t2-><s4#Q}k1UHT8|fGGP>cR9iXJ&cag1eFC&3^opl-%k;8vN7FTQ5vrpi
zhBl#UrDwTQ;358+@hy9y@EOvAXiC=A9S}22J2DsfnzJ8phOHg(vo(o~$SCm*>&%=Q
zy%ek&tHf>AUkARMqC$UeLAW_4BQ6UI3}YC(rjg+c(OUfE|LNSMd!ySTJm+eKkB0n2
zxlY%gGErI~SVQ=aFrXpJbK$S>q0o4hod-<u@{xKysuTBG{&h7L8;5)8I5D2+=-*|Y
zX_*8q41UwSaB!$M_KK@w9udAltP5X^Bnb2E9niO?!`f#=W5s1+u6Cngp7t!UiKmT=
z!qqYshpL)-3rkGLV<}lZbWnjpyd7N?*};Lr(`YOIWO#wl-rNtJhaT4Y%^koWy49hc
z%m!~4Qv%n5duXwS4r}K`?u#|JY@vVXo|vlr!Nb}nL^4*1|KM0;K1kGLJ{m?xhjK76
z1L@0MRt<}6(wc~MOd0x;^ILZjAA(My3Q-2k@yV#p{`J0mz61YNCku2JH=FkACb<_A
zM-A12q~(01E8otLWawnFhc8$M3T-rl_~GX5ksD02l3aE%<>k(Y_7YEx(~W;LGVZi@
zkvJSl;O}W)@?*0;1;fE$@F}q(y205!JX3$#qO+VZinacOZ{S~vYbBjyBlv*<tr5s~
zA!_tquq7Y09Mo5Hk&$d}0QN|6OMBS#K3s_#6&%iu)-=G6Gu!dnctv!GZbj*q=qR6w
zA!)1bj?coPe$ZC4oW3u20?Z7wp6aykMM?$lFW@`$o-xDUkgCOfp?O#a*HW_$i4^|S
ztfJFczj6bbPImW9&zr(6qf7DIa3#-5vKCecDWU750i*%`-2FA*!8X8X)~txaZ!{ki
z_m~Z6HKe^GJNOVN3l61UVFliJ>KB%-7!KIjUEmzo6y#TGyq}bQ&RnOM!mff7;p)DP
z@HI3E9O?0~v1F&a%yCnmf^S;4dUKNVrE@T~7;5U83BUIwrp+(8p*?`jOLqBgQ;WdK
z$Zg+B>jTik?A6TlT?KY9n|!DJCNHe)tGdSAm#n6G5J%9csx>-ScNgij;j(x@+80Tt
z*F%fcOOzcl-$K2(yrMdw1sw-Z)?}z(u@5yLa3AuMZJtd462&m0uX;Y@LJBf|`vm_x
zbOU--!>3e0TwXi5yWpZ@C}Z-sPXEH5D4fFBoiCg@();Wa!!G6+oPiyuzVi)`0C>v@
z6Nl_^$U?RY{*Y{?TxrV&7s@=Y&Rni?Y2Io|>3&7J9L)hcbB6A!8R1>08A|?57O}lE
zX1OOJJM5qE0zAX{Cv!PkOOow6<!V#12rwxBcw1B3S2wZML3d;>z{{wnLcF(<?>{n5
zlZQUyj(K|m-O#%}it(gpmi&<Ba}$BF7_MnXPE72}4)QHS_95}M36in!IwlpKt*C@s
zJRcM-sFuJhf>9PZtIBp4n~+!D(eA}g#{Eo+YAE%#SSUXkwxoY8dF<AZ6Z{#<@vJVt
zDf0~J<rw9C>5W5PTX{{2?=>>eH8uaHI~9G0u3~a!T{G)yP9hNU6$(%<Q*#BWZlqIL
zaDIK9^e8yh--n&&i(@t-2NZ|=H=%XdGq9g=Idyn_&^?D8q5W-t3&{AN(&ehBTvzpA
zGy~a)_V@HqCI>n}V=;s8Cz<YU4Xs4x14Ai^T5v6=E23d1rx`*ul}tcj?kYH${^B|9
z{hc<&m#lWFCUJu$Qs)W0Ik`5}MAg^-$bOKmgWP7dfWqA;(4KplnND3R$U(ILR9N8O
zk3Lw{UpYwLe9aIl<{hKXR`ppkoWA3l<qwjTu!v_kHkw&2JL0*=hE+d+J5B(shAOD3
z{?UOTeQ(Q7rm~-cp2BaDXXtbGh;)*)VD0AMV#QeM2^FEH!I|lMsLMVpV`Aj?O~_cs
zT@CE($A%?mATF|>O{Kk=FxD1wF+o>}>W=D!;$O{t@+^KE>g;U-_aW!Pqtn)9e5t)j
zktZ$#mH1Y*SN5)Gyr!R*rdon2sFXS{9mvjOoo+7^pwf{Q9?}1;sJps5xEfL*7tr?w
z{gAsrLq)fOHN*zQgeif!cpG;+NtSG+Z;R@S=e(y0J`|e;8{E6mfcq`IhWg0fRb{Db
zIWRPye2v^kTBf(m2p|Jh$2N{uRCcW`RFYNL$-qPDb5tp7ys1!`L}de8OHs%SJV*4z
zF#i^8AaIFXqvrfl2n}>swt=@hCBPeOy=s)_2(uL_^F4%3Xnl4R3jurBwM<V@MvLBK
zj&=BTv`OB2cULTxn_bY2Yrs_WU6lZ}#wkZKbr=)7-j$;m9IU1No7>=`Z4kX2V1e;z
zLEC@W@D;6rdvsIsSCRzI<}$#m^d=Yty=r}Gb9RZRs^_1;4CFgqk<{oDm~?Gbm>~_&
z7&_ly4{@^>nQ{05wSig1=#;Iz7VrkWRK?i|*A<mVc8~3(QlJC8TTs?<&6(!v=l&qg
zWonT>pxaOim9JK%AgYlixJ}{H<X4o<aW+J@s*9-e%uM<MdlLx4?G!Iui_?FBy*#A0
zMWDK(redyo1=b;W*qwwg=48k(_Ox<spuV!DyQO~%#G%hh6#mqblTcM)UEsGTJ}?2U
zTd4||0oPG3XBWzM1;6ntkZHboa0K{GozZM`4|0Fa$b<#-0Ck>ym&~(+I0^nbCE!d&
zp0a}#ZCE8gC*`;Ow%5#jA`#^mWozG1?@zD9F^IBg`>P_r7od-<4Hre{b6x#+Kyu4E
z=QiLZvZbIBYnB#j6Y1*ZdhQk)@=io5DC^p<lBv=W{^o@l;5Ba_N{$T-9YH%OCXu=9
z5rydYYtF)DdfxdCxu&k55%F~^u0)mIF<4jIJ8HG6sVy5nN==k~mA8<;-4w-cP^+m@
z=cdqeWS_sP^Dy60I=~fR4?7-_=a9CRDuflRj&=bKmZL>SMk&8kJRWTVS((Q`S5JLK
zB?QBFY7Z$4=w1@Rj9fqO_mZ}%g&CE-EdxAu(>aT)i6kqsN;<KhknvC*{3%yA<pE@q
z-69^L2bF`c)pTy=JJ2LMhE`BkWj>?lfjnTYdm7sY02HG<f7PxhpI~^*+{ry;zn~4H
z4V1lNo#~Twk|Lw_YvothFStSCPS;Oc88RX8NU;i)#YUyyL}kjWo(xy*lD7CI-~_cK
zFoS6hE-!lIzg==I(9kh^(-1I=9OJYiOYO;2C);=5eQb#TZ)gF4fH%uCavpe$Ml|W}
zBJCY?AN>!q+@0)-x_bIL5-T))6a-p^jFKzRbAjX1A5b>Ap4750*gW4C`k^;~XaW=A
z&fc!nXY_7zcUd(>8_x^XFlk$;E7qGn2oYY?*PogJ!(dx>5_Hj*hvs_EJFmbsphx-V
zRePWmg&-+a-cg?5<_G4oqcoG4znHJU#p2q?8=pY$ayjI4JS(}{*yN<ANPn=KzbAPe
z@NiX;%A`_r2>U`0$5w#{<(cR!a)fFq)e&CF#A_Pc4terqUREMahwu7f=7BpOXwM#m
znhEPm@_}Kh?)JgCVPzIO7j)1A(LA3D=}onSdvXN$g|Wh$l-bBsuUCE^&K1TYVJM?G
znMo*~zv*l!%U8jB7@)|T=u~`3+S;NnHpp3ygdU%xd-)CZ$4ix?#4im%UGw&0>p5SP
zY}=UY7>v9q`PWwstphw%j&xSytg^8-g52Z|pm!yImAu6I`>wg;(59YINpFdLV1M)+
z`9d*OJy+f?@Ip0<Yy_Jzz_!Kv*6sAI^ZspO{dQMJWSi58-Bm6{$2xy#T0xD#Ht2W0
zIg;TsNnT5?G4rIYR0K2;wTHUccVOMASxiGfO6z^=rO8MgYHUD;Jn-#i>Uk$4pTH*G
zH)K!x4wBAT!OLKKN(yg<jL3CoYo@MTPV&m>&`F4Z@423_3rH4ugWYqVW^`aj;G}x3
zs;TxSxzD|WO_l4xH*9ZArb#0c>3gsS`HrkbzA}w8r>HyL{<a+7dHXVr73l?^ROEs`
zpxNX+aFo7*Dz~5#b4@|{*Lzk$6=Z7iu`8cCEd5-*lMz`Wv3u5fjw=heJbeSSLP<|D
z)4#i%dpV?O>~{Izt^_zioJKET8h|5^<`jcf@LuruaN?f*p?|%vu=&!vp4pt>`3%)z
z5}7HCQF8;`;A`T41C4=`U?r&aDG{qmjr^CVr&~fx8L4YFwaeKMdzJqM=nU(fd&%Bd
zg5?*xTWumYY97;ndp|HUsZ^IoUKTfBS_52w=&&|y0ooq;OxE^gGFtZ`INdQCoS9lk
zCNOfkn-a&ypfwa(&>|><-2*jOx|tj9&Wzq0N;zNja?2CN4gWfHnabom%cN4J8|~>7
z_XIqJ;hre46^Z9GXrI6pcrdPX-%={vuu6~@!y|!Ozyjz2;<4?9%iKfAIN&?-oUG?-
zUu!x2ij)-{E%-#vqAMxuubjqp_I<J6$ZX=?!(O)eJ%<%d{AbheGdmQ|p|;+Ba53Pe
ze)&c+Yn)Ol&)J*%?#{9Ig!iehXFM+G?i)|N0RB~sF6XC3=yT?!f&xx43Un&m*ge-f
zlzFT0K#Lf0lRKK_-M|28Cz%<l;cR8*Hg#D2ixJbR<ks|$Nh$&hnaA$lnj-9b)<STa
z+aWvbo02;gNq5bl1uO{fqPwNdQiXLU<Z7UX>kZQ{c@ox%4Qq}wGIj_s*)~8@nI#Gb
zZTUjqqRvWm+&Qj-x{LP-cY#?1>3v&V`LqDBu5R9%bbUjAW@=_%O|g1naetc`?Bs}Z
zRduJpApV%laE%jQW;{S|!UlX9Jm1$F>O{ron~`W{FK>O{2lYep3sg1Q(n|r;ly3tZ
z<0y$amyj!<o`_Cm18c<)EiZq?H4h}<FPtBk8(VlMuZl3|!1W3O-wYHnhiuod&OkG7
zZDdYyIqQLDW%Ob^{*&lz_KyFsX8GnnR9B8mk1II^JuG+ESa>>rl6j-KhiAc+fsDv*
ze@8~*rZj`_Oz5!8%Jd^>7S>!QD`Z^-mipUC6s`%NL?Q{q@HIA4w~PsJndl;a8oNPL
z4}Sy2DO;n*0vIxpctsZ#l!EEN8P{1Ro<Epgfk{DTq4V-rkb4je>~1@Z9-slTAk|QO
z*!>8e3svw>!IpV3dpFNW=hRS~Y?7lPRghdA8B#LJSYBx3zO(D(-90hnEC;jmi~ja)
zaJ`0WI=>)l+*CaaD=1ePyAj4KcN^KS)LZ={ef7X|B?Zq{9M&kn-|9YYr(*(k9qi@a
zZJ$WB_Z7*!0g0yqkQ*H19;mS)tu-jp5*wXyANFz$wYPj0E#lftZ6n_*uY029{O68*
zS8{QA&&X@Xqm+~MDODZqA$E_V6>xzqgP$RtQA6OBYX-j0k03445+qPVp?T?EN3BcQ
zME1Zp!Bw^YFoU#a=W1{$xTByQQimO=x&>ANs~7ewnTO;_cB+%myUYpnmADq2h^I;(
zN>!Q@{!aAQzyQ?5?2*l?x!En%9z**Aqh%8?ou-{v103*A7IT5Ywkpun;6C66drV!2
zx}%;hf8e#iJA9kzr|PP;uaTFb05UW$2Va=EirFr#LN>z>)d!jQyz-ckL+K2Uqu@RK
z%Q=xr!q02otGjS|<h5rgFqmmYshA-7hiQ*bk@o<4>4TaGNaC%jFf{~7<c5<&ozKYT
z__Bl=xB`Ko#i}wWh&87FMP%6DAVjL5bog(jhTO_x-a4A|_K54fya_d*8HRVZ{~-rK
zm!V}fR;GPd-}HS&vgt~a@z_o%9SX`iDjuiQa8`zvfzNzbQ4KNz%wu=6o8ei|DiHI9
zT&%4c>O)zwBCt+brrybI@Y$V9;J(_q#ufflx?g5%a=B8XsE^=iTp*t5;;YKgsNnag
z-?9H>v<EN>XGvH~-QD5<pTMW+j?fI+hOTzD^-7g*fv!#&buN7s=$0>L<&@0(Syf5m
zV}2+GNLpzIK|cUH)dz}#d)c;lnJSS!B6)=_m5hO3Rj<!10-m|%dwZ+Cs=L5qpdu7>
z=cy9WE6!@=H^m5YW=UhHp8Faz-@6>s$lkECxPJgGlmt}sS!n_IP3C%^sT;xfkRj+y
zwuG9b-Ubc_-!lv8&6z0Nkg1}6Ue2nQXRP%nF(2r2%n)Z!haGsW`WI`M(uujD>_wLN
zT9Eyz#RXG53njV0SXnK=Mjmq2lP*CzQhmYc9)UJ1>jU@U`OrVi6NQffAvAycdb(&T
zcmo+uRo~oO!9cInCii$v6{Z)v+ue~&1S)CXV~1q(ysf=s=m_(O>8$RFtYeQd6C_G{
zBj|FxB=3>yGG9>zs(!A_^mk-8?<L1-$_mu?47NdDSy5;5wtB8=e%^P)EtRv`{v#W@
zLVeIf;FFE#JT_O!mKjiMS$k@TuZep)@<3AySe>vsy%%*Bs}8lMBq<M_IyjT+PR7_f
z6t02cuH{*`S5{jV_FpF}lH-s|ipz8>;H9rt`c<yJ(~S;P{dC7sZM~zQ$?o3FYSvxu
zRu!s7Ld}tduE}JE=bCbu>ZVGL$e=>F1qCT8`B0!TDXn!LDsZnw50Mp^WRF*|OesaC
zW-e#Et{=4@s||{HrjLE5$Lji7LV~ZA8RTuQHnQB146Sfqwb2<&?Kk<eU83X%@EZxz
zzg-c>_mm>HOJ!zvkvXd71wohx;*r}bSaBGbL=L8|xh9eldn;xRRm&H^LdZGWS!N5f
zQu>3;loz5(<Tdlfvz;U$E&Hb+H(_-i15{-#Y(3DQruImXmzw3}IC7k>qihdPXD!q$
zaEpJE^p^5B=~mDMQ&gY4vi!*o)s`;oFlZWB0*y(j%Zx*NtL{<Tq)YRrc$}UjjN~3L
zdu44gm1~v?E3Dv|4A+N5@Bp+)SsOWk+;E>$J|wN+A@v!D5vbu>2VJM;I@;9Ew11=P
z<yK<Hd;7XJATw!zz2@wJpbX_UAgkGU<wC_8WF#;hXht_E{vuoAgjVTF`eHkP29n2A
zD>X=w$a@sEbs`~mC)BupE7-JPQ2IRZqkGoc7VbwjtADK*c9GQebcnu6)^lKNJS)0>
zNGiLxDnFwU%|PljFkUs2+Ly7|*+M45HFJAHS0o|dUy_5iSNW6|FIO?cXaz8V`G}2B
z9JQw@swg(0Pm2un3ld!`qD5RPm0GUMC0kygeZ8X?HGS6gG^4L_RboA=0<z1U!*t1r
zV}81}OExickfD$xWuB%Gn&F*;WTO+*Jv5b>?cT-YKt+xRljLb?`|7YSblsHQpg8Yp
zuo##pW7Yo@YSF4{g?hawz1$Cs*;~k_n<qmTT*r{>*dnS5_Y*b*T9eN+EPjbxfCs9E
zD!o)Z@Faa#iM(h&+fR9~`XNQBbD#1FIA%=``Y?D!lJ6MczA6XFQ4+YgGgL;^k=J&1
z1(&#nE1OWUk|FYyzJVT2lcIT7RO+yLJGhp@CwvFQaqwJk7wjl&@@%2@ky-2?mjN44
z5!l!W-7T|e_rsskp2(JZ4`_EXg9AGGVSD5B`sj0*bmOkQl3wiXlm<xN7Om=EBwj|l
zN#F~38-W-hdpDHjT4^{(c5uuA($%x*s;n0KqY0_sf+N-Ot|O9WG!4vj+S!5fjTCI3
z!4yC_^h|gQQ(nbmcY^bgsq`^sVWyNywhhD&lV21`Y(Mp)f<^LPiu;gg-%=z6|B<Dk
z-eM(P$)*A3YF5)7RnM{6B|%LmGUg1~w<%K#JIIC>^uzlpYA_9zeOz6Wb+RI6lIu8G
zT#iT&SOZcFFu*=)2e?3aQD4(LoSncfP^T9S^nLW@1Dm05-rD4<nv%TkXlss={RN*W
zj`<eC-`OU<dvHIyoM}{kGTztL(#gthm_BnMG#OAkyO^Kn*HGtzr>JYtYjTq8pVZMv
zD!3IKBh{6;RrA0z-m%~??>)~iCdKy7U0u0YnTX^uar8X*Jz%OiRq>1MtWh%KK}n6j
zumrR^eUq;1tb%mbc7diNAHD6MMnSo`Y4IcXZ?Xf^UYX4{rT=29QTvvSO?OqvgC;|D
z(OU#fU7^an|I%Nz<4T&63quZ|3scoQAM|CKGQz+L(v<$q*8w=ErhWTeOB_wHUh*wU
zW%6`24!D#~asXw;RjC=sShUPr0QRTVs)%hq33=sty-Ut`Er1+If<{BVf*ByAusF|n
zd&<>xvAQmk&z@%O3OP8~e#o<9^JiC%X98A>{F1>Zb;#WSbyIIf7UmBkcX;&$3)pzX
zpSpj`A2d~6G2=a(>H3b2^qh7Z9S6y-?j^pk?=XmNN%2G!2aq^$u#A#-O8f0h-0%b%
z&AnF<?wQ^gGfYm{mQk(!LtP=VH?>+e$s-^xI*%U9B%~#j_8?n3%7C@DR;n?sUK?d(
zeI%p-!0J@CR)tM<-%<l5gR#fTMCEJGq2lH-9GZ>w1OVsXnQM^8c>^UNI95s3dX60P
zZgW2Z?9vvB{_4ff_R1^d5!nwQgsqY8)y!5l1t%f()PH&F8EViOnm+Eza5H9qVP|zd
z8zN=YE>eSD%*cZuDISBh+>Klt96Nmtv2V`mXiEj=NiI38d{@$=U@>?v6Q%n|%9suA
zpM}4u>X~uC2k?Y+7o08aEcApsmo=nRDJcAbIKpnm3DsB%#~!*9z!d*jI0HFKHb5?8
zWmFIMF8Tuck$sW*1G>Zh!#?+X^}y`&@H%M=%0rD)rKhDzHy7<fe*4CfbpxBpjs8A)
z{n8r)PpKO05#I=`Z}C=!;}$Zno$=^9#W3Uu`7m`inSk_iZFV0hP&+eh_d$=ZikwS*
z=)49Fa>&vLsxLVjxW4!)prZRM^-zOSS}m%sX`cm-DOyCeNxMVOvdss(s^7ZSGp}5;
znAX@z<W%vvWCOLqmE@Lqd&{r8ANZOh5%Pd-2msTYlqL_rd@o9t{%|yx&!+}S$I1T!
z+K{*GZ<TS(VYo3co%PW(9VGb)YK0&&iZo{^)X)40T90>jk+7i^l8<x<*~ufCl~hzw
z*NXsq*)sA8*@9lA-V2<SX0YSaT0(Q3HObn_hinXu6c~UpKre-zyaF8e-}D(lfIfqs
zc7KHLAmNl0$tGDJWg?#Bs6#GwK?<CM@x!a@xns_kE&={VJGo!<O!r_m7x>FNCt;F&
zDRZ7qB@vkd$^`y(CnEKL6W~eiD!kP<x?n3fT%M5pNgc0P?3}}Xb&kU)D<=fcU=^`X
z-cwGQI^@iDvYuXK)BKqHgtNO3#?yQwAq(=wd(gF!>Pc?W`2CO2t?)x-2pNjJ$uDx>
zj%G7|;BcOtJfgU4ABxQ=xl+<8<u~YbJ%se0y}({q(|{D|P5(zt34!Ed_f;w%t%JDH
z1(p)P3BN{LxQC(*k=u@bz>boet_<yL_A-JM46)sz<FQ$I*CI7u$@v|PF}oSuTLYP9
z+pI>p;pjU#;t!>7@a7c!qFbTk(JWNyd%+Ba4j^GgQ6T{&RWNKiq`c#P1XV|B(wSQp
zz_Y-0nO~toDx&yE*`)25oeGwaxpEcn+=b|0aJ$09?WOc#)mEwwZ38z@xzL}QlU#Zr
z2<;TF0)V_`Srn2WpXfg1V4!AscE8JL$>cKMyd521Ga}wOn*-cr%^BBQ#X{wt!Y@E`
zpxGvbx=`*})MeiSMtNJejD(QK@ENZY=)ttou0#slJK<5Pjp|j@MyiuB5t>NO0~&Zt
zs$CG}UuV-x?zmlaJGO;S%{*9NN7)JW1mfk#*iX*p&@`%y)n}gYoi2A}dKEVZ637~w
z1Uf;on|_GaLaw+i@O(cC&-0&A{|8P(u9b|WwO~VXFCv3)yT{Po7zyLVhv*8JQMSs?
z<$%Cmb^3)@l4kC-Kq9_J>GLQw>Fg~^k31s#!>8Q~<QP_1vKmTIK88KzZud6qSy3gh
zgVzI%N}A{DA^qr*DNiX1<-BbZUGC4)H59$dIyx_+8{tH<qiayPOTN^uS0sSt@)-W^
z*{=8xz?~0hDR|ejTh_v*a4ueZBwEGO*)yA+Pm-QQ={9;~xhM9WTtJp0ZI#UwAeE6<
zR9wBdnd`f4uxl)}EADb$g7-0TA~nkVbQh&o_w<##mktGwdAejAc0{PNwe9q7KZh=K
zb*GoB+aZ^!<F2|$YPkmKhdyCi$}buVGgm0bxf;j|l4#(T_pA*q=!^nrO|UCgm6R}c
zYAhaHf5N*$c2K1T6J4(W8FL3w1Jhu;`f{L5Q8oFV>Xm#ZlZ#kWo&dF#J-MlDerC;*
zvD(_UGTD2#3(5jgCH0t|XcAY)yEdOtA1zr~9u?j*4tyX_gYuLrS1LS82|xk(mD-~)
zr}Xy1SjY6%FelacZ%cLpi1e8|-5XW7ko&HKB~i@2IgMFEHA4>o2bo*QcZHg%3rq#e
z)cwGw9yj?&1v=uDXKZSw0#k||Qs^NgSPMOxOG|CYOifm9t?K8=f8Z1*6L^VKM=Da^
zfsgKq%v5$J_&`B8`}!ua3mtP9w|+6OlQFt3(IR=(yHsH%lU-)A47|yYpsO=M)e>-;
z?KFCA%}#$eN0w`zuY~GZ6!tVO&{3p&h_Z-T%KYaZfzP*3@;9PtWuAA8ELS7D=~J*n
zzD-fr*@SwoRwTcNE3ns-0B}jh1nh5jTZhKA5mXZzrf1=@q8ExcPE-Ca=1I+)#aiS_
z$zJe=A`Uzb;p(Q)QcTY7VZOOWu)~18wui`e(n!5sGtSKbU6BvSsr)10AfSbR5xm~{
zg!&8Vk+GhBCAlX1o3g8X<R54gYa{a*xp%efv1+itCa8mD&bep|`Q+~auJ`-WNXSP@
z&;S)inig~i20MB8Ge^UMLe&DSrQ{qnzSdFf9Myx)v~%owvcId?`!&;tY<DE7p168a
zj}_CL@!lzAjOr=%8LG!{>hm(T-0>Yo9>(4x4b~o~ak2$?%-P63byF^PTe(>+Dmp`b
z(@M#sOe3g)!UuF_qMpZeXSt%Jzw?&+A#{+Q;JHB7lJp{XmrU}O*@rL)TcFh1uR;Bo
zo@5{2ENY<SlVTQhxcG<XfO3fYw(Ad^0Jh6oLv{mNNt3BF@=3^S<znwD+zPI+e^%TE
z!<v@1=^@zD)C<dwW1T|rcp>x3vl-l8(p=?c%BdRli~0c7P2I-mC|=cyIYC}<UP2RS
z#CA@;wlIvYWRE*{5bqe29D`_WKT<kC3PwlHE$U7CfD4+*_A$14e5L{eW~ydn?m;{G
z)ad*CjfG3$ZfVQ?@0hDloT{<npVCFn%E(Qmlw<he;CSbR^7uQ{BV-}jzk(b3Hxs5Z
z-Othb%t}vF;HRtt{he9qdTINvZdRh=9EzZ3PHr(dLfx2IT&%+~;kV2;Uq|2zU2W4A
z_JQJwvI+e`)z`KHT}HhF+S_vc2GtYfI64J60JaawxdFg4=VJ7|23Jq<{$g{rer7Px
z&ehBkb$6o&qcfd7;8)~spF*=KxdnEGZckxI6Y>{xPuUHzDcU=RsiRy)q_$Sh%3VBF
zDW$x&Ldk`m?l$BX@3SrSpslJb#T{guHy5hNETap_TEKH>XS4(Mj)`EqnD%7d!p%T_
z{zgw*#Z2c6C=s0xquLh8uae2kTid_s9u4cuu8v9mB8%iLidM<><s87}oCP0OZ6)7H
zFEPn%cPgXg7dura4VXM>@bQwQa*g|=sG<A~^hP}hoK<+iF^Ss8evnu99aDTsfPGDz
zbI??{KU5d`t{n$FRsKiyg8<ife5osdET^8LPn|F9JMEk@4j6@>_t}6#=@q}L@DZkW
zj8%@MRa7UWMXis>L|adCA^x3RNVZ2_;z>1B$W6r`AhY&ovUBoM=so(%a~Uq&oRF81
zyrIMl_oezMhsn;sKKh&FIIvYZLVmP%Rd+UUDymTqg43jXWeuP&p7s8F>^@{wNroi9
zT!mCuUO+pM{oHL;cjaH5Gl4k&P_ZkpgMNc1v#EfI%uM>=6%Z5BlW7TF^1X0H<rArq
zRVt{IK)ii}_5?DOe9bu|d(a0PQB{SaV+vYnueL|>H)uldCe|ywi3ZS3$Pd2}Y)U?L
zHsI#jr_=kv=IkFthYWL(umJv0E<lev2<0&93Y*0QJhhkuu-X3zqY}>nfOmcJOZbxf
zW8zNql%iJQch~_oW>W0CWoNwCbp4fWTmx)dfJIcMHs-A8K0`K8ZS*WI$X2YQPh$PN
zN4H$}6oMFHaWJkEY-^^CjQ}Qjis2=yneKnQ9msez&hx@o=KaU{J?#<P7x<an5j{Y5
z0-t#*$g}c>0;3Vcu!$Z;bzpX`e6FnMsiy3T)@A#sG_0J}ATyZdXctfW<SDglmAe*u
z0f@}40J_eA@xIG&dsmU{AU%~<D-NMQ$p+NEVn(sUxr^H1$z{688oO7rozrVAFOh9k
zv@CvGbON}ap6uJCN>+}Be&pW-PkBcdw^3j6>{opO26}_(>)hodO#TG;oC9=Y$oUeD
zyS4h6$K_5#jwZLEHH;FOD^2iv*ukoMlA%tglcP$27%P*jb8{dW(!J2=zv@0;b177_
z_6n*f^P6l1G>j}%R#9ACneJ)M7I{?kf6PHudG}2EVW=azKka={1!r!?-*h#*N?8ad
zk$v4mOZq4(0M)56o+r*9KquEnfJo_(TOc{3SONU2#J%k$64&IsE((+MK5UU!cNenT
zrN+uFG9IZHI&$G|zHam!;IwOIfggE?oXkJuO7QGp&KCABXJDs*^L8Cm6RzPYvhCdB
z1Xg8yRrYYBz%b?ZlG=0?^-h^B{l0DP=JApmt~mQC#}nXQ&8{0h0xQ)=J@457*&BKY
z)-N&Bw!#n83^@eXB)hpAF$c*#@KTh?E2NH-7d$4YOtsg2K@}*t<uRvg-RDVp$r@lR
zj1`27*1PqI$GEDt48<thA7>A06f!ZtqoRjbK{Lws#XmOMohPZj@C}#JO;HKp4)-Ns
zDsTqwqa5Ux<gNlTmQ({~fJc-AC70L|puCSsX|;QSzVI=st#qz;-NqHlBl1kASKWln
zX10<qoKrz9Ro^j`nyfkuJaHURJ%Ml9E3v7{{`zxJoTsks58zCn50*;5RIy1PB4ysH
zWHoYcW-ID#a!z^9Wo9Ni2{>uXN2MBs$~}zQbO&;fv?6d(-qCeJ_MK^o_JCWt7_uqT
z%V|kF0!~3iDB1Ul>f?E<c<OzpsLA%A@23=aDpE^SN79F{`3)VzPLhYgG2U15F-eK!
z7Fic^AKf)Q4e6{dU^Z_?6<cWq?I%}*)4V^xZ9ql!S;Zyye53_gS@xfN4lpH^R80Wi
zdh2^)sq-LMb=uL=RU7IG9+O;@%v2%vv49k+=e<NOr!JC(syg=Jz$m~av(Yc*0+pmp
z^KvfUwL`MbjlqfD%@j%xm45fcQ@Rx^y~7kW;a*Sz0H>X(J%qjHso|RF8k)B_T>~sq
zH1)d5tJpn(&%jo4fp=`JVt@y-3QmFBrS%l;k*3ZV+1-Kr=gVFdUnIL^c<3q_4{1OJ
zu$$eB)SvU+K#6KOg3%Gs2Gn<F6n>|NQ1{eJy&m8}Zbz>C+>^UJ=jcJof|UEt_UdHM
zBjrV){kjM=h@7FwQ+!f)bpz7M=pU9UyrDQ=T||>hFKavadj_MpF<Pl?p2_KZY-q}t
z&n|}QFo51{fCFi|3zl2PUdU#_A}tbE6H~)Oga66nbhE^h#OBy@j@2g#74bHPyOvKS
zsO5{LQd+U`j;Szsl_IR6pf=hgblz_g7P0ko3w1jcR|Ts%Pf;e^iHwT1*YC$yS;UO5
z{sr0Xy;*#B;~rsfb}MAOzLMT;u9MTBS%`E8vvTNQZ{Mx5y6lzkdtoVGBRbPk$GTn2
z5jK%yg^{sZ?vuLYSP5`ngrj(!!`3IrTE2~VfQ<8|YNzgSnwpnsgeGEP{wRKZxi5Q_
zuVwsfWFkguso0nOZT=VeXwNEJz*P^LL_ai`@2o!)_z=5e9-!lNYE>5>Yn{TU6H~Kl
z=}+rbfhaS;@|SUN&PyF{u$XL$2iimY8SN+YKUp4P0eUEBt-<A&1;^4aV%4(x63L}Z
z^n>v)KOX6!Ukuv$D%oS)%Zw84L3Y0~U+GlgBX(62Z&q-O1>wKZjkGa!lJ+T+BW~AC
zFH1MC%WfY!r2nSX3zsb^v0b`FvCq~v&R=|-B~I}OLU6=*iu-1fMv}24OR{AV`pWpt
z>@97^EhC;#Tm4hfdG^gJ(cIbk$eigIS4zj;i(J+`>*46ez)o?MDIVV-p76)vzXheS
zG=k|4Mry?3(R(~$+Mn%+RK|<V>rDg2MEIOpPxR6`Of{iFqT2EqZ)NCXA;e@u>)P!C
zv-KS{!->^=T|@iqIou**JK-};5JI948zP<*BSKI=!B8)AV>T<~SOoKf;2HCmz^>p?
zOKSN9>k(qLwx;@m_#0S*9Sv#fDD~GdxNLyf-C(i0%Vw13<0F07BlW0D`d2!R{}~!+
zN{ObJv+EuPc7vBAE28JKdvPnn{|T^o9iKqd<{n_zOxvy7m{a0(VXLJsaVOZ?Ff&>;
zt7&8^H{a;iL^#OcwGNI=ur|}$Vn5Ah>=V&E_K(<<knqYF&gTel^i}p0%X*6~l2&%r
zAS727SMZN`nW2%G8M$0mQ=B6-5x$G{jjKvuluq>3j+EEo%m24aWUBFSpsC?7z9iJ!
zvYkJwJA(z8uewh99L=H#%M8=1#reAK!j13(Ly6FZmI$xSXsjbv)7v=vG?8ciYWgos
z<8#aBFqrU~VvGA<{)3<i_t4h`KJb-eKY3oK=9dcJ!u`2}A`pHX$lclxy&*p1FIw_V
zRZ53jmx@YeC1Z~rD4(iW65SDd5cY6Gq9&X$%??J<-`oE3wPBWo25Jl7ZA1=`p`r{s
zjD51Enm&b_(h=92@MrNiLPZ*v!C8~Jr@r5a(msXTLjZ=Vn)<HOL~&LR{|c;`b(7{q
zOlsU0XkWVDU){74eZbqZ|3}eT2BxjPeR%HDu0|;mAb~i6%G}-E-QC^SY3E(HIuam2
z5|R*T=f0h}`_5d>+};0s^M2<8geOm)-1qOg?uPh?)#XCXa8W!P{3tB;?)5E{KU=rS
zwcIuF{E}(%!ax#wSL%$mBsZ1(U@GykT&4*X{-jPQw^|xq_nP^1YDk(ZBQT6ljb;!l
z@!H~-=q==zxkGA>Ethx@Nw>vNHsPpTXun9TfP0Al;D_WXa<7<S)_{0?*N4<{;TYSi
zv<(n~*hAm(T_72DZ%>barS5Qiu(!p5LBVvvNzP!V>{w7Y*1Tpz8`xppg;%mSCpuVG
zY8T?$gBwa5mYZ_p(68thk??`sR4R;Tilf-aRHF#G-4*<cmqx2&2-TkGCUp;~;1Mt@
zheONQ)uj(1<-aF?QU-ageXD<=!VLEh_d{k2y-G{DJ(dH6a{bb}LwCcw#Oj=eIflK5
zkHC!C5-wza;f>h}w%ufNs-f+Qqm<|siLw*eB)KXnl&leUr1pe5kz>GrDJf#IY49AW
z1{NbXBpTSpK>tLZMivQ|;5ybBWFLQoT3GU%*#~vS|7ZPzH4LcIDocGx50qim;y3WK
z)O2gHXLNLG!NIVP--bTN_XIwt$rdWIJ~Cb?!*s{D!`&^Ff-T^TNSI3VD;XnvRN7DR
z6E!kzvrsEqjNLZlk+y~-k>`RvVz#O%25)V1lgVs0;fe-`|D2e-K;DWs3pHn~k|mN7
zIxjXX>Prn4QcK?1+L-3Cqr^#a1>#RwOFkmI1aE|J;zh|2;lI)aCD*IZq`Ko8esZ`5
zw#CX4lh~7yI%K}}K5Rg{OXc~2*km?4QZe)oE8<h=<<{ZOM0->FHTgek0@fwEGMK|H
zm;5FDu>(X<6E*IP*)KY&P-I(OM)_1r_66k)^}%2~Bjz}L)YdXQm+2j8M-Ir`hnQJP
z9HkN2()8!>3S_(=u&#=h;bD12dV|#c(oM=9?Gc!-`W9(#txfLc9KkN!D&fZV`9!8|
zUfNIa8mgmy!eh|0a8<EM>Qvz%Svy=#o=GK=U#PHrGBiE?EZ049ROn63w6B7&9Sx#p
zW`nI)q%4vqE~jEqQQ9ipXFHic2&37-MePY=TEj?Zv9D|~UrVnnb&(q>RI}T6m-28H
zVhCQ6=CO_py(odjih^KGuuYA0#wH033pjX2q~iKJ;qjzw|3~~8Y0fRddSWK)aG(a>
zP5;&%5lCAiK2!YbS&3^YgE<5KE6$X=Q**?<c+==@a)P-779W1CaCoHxC3Hhxg-(@R
zqvne9(y!Y8XUoJoa%TmnCoWRfI5O=9JyW<C*bzA<d@lVfj1<eJbs^6Ar`UE$2Ju6<
zr@ToXCDnD!7Wc?l8cR`l38*doS6Y`=#{Qq!SL{U0<X54?DFb1T9KowZ#*jVOA|CTi
ziar%n@UrC9(&>^NdI~2{%Jq@%Mcle%@{!ol^OZDNlc;oW6A@JW9M5eFI4_n=Y~Z$F
z#uC5=nZ4{kk#dphB15&|*V?AC?c@afq2(uAMK%zt<?^=u5m#^t+{`?b7#{AvbFA<<
z(t+@XN2eZ;ss`eSFH#q3L~6B2E;8EMOUSe!_9@XX3Uj_x7!f&5mm#)Nv2qDo%^xRc
zD9`2Usj;bh@wV!l6e4$L6`V;<mo7kUpjpi9(7)DPsS{D2ufsy6(*nC}<4Rr?pNr%f
zN8rDp(Wy&x&yW*RN+=P6h-bDr;mcd*MV`p+)Om6o5}!H-G(ufdi|F<te`F^1i}xX<
zL2bJ$#+I%$#OiMPH<t8<3E}~n$o;^KeuvxuJ48+qTAA-8R0ua^S1J769`>Ac&svdE
zi6KiL+m=w3nk+oIWT6=A9&N774YDNJM@ma61}i}KM~8~%!`sjtGSAk@rn4^+*D^aT
z=dEvXH{8MaANWyzBy6`VOaCg()zzT(J1^4#s#tz3w-VI!KVpCQ0{o0L5fj)n>ZtTw
zo+nqdwwHf!S(bF1Ay&iINE^N<SwTjSfh%@~e!3uepY%92-u}+oh3Y5HfqED216}qa
z@(WVg)>XO{%7W`L{{)?KrNC&h5pzQvNY+dK8a;${CXYuS@LLfFc{Y5f1W85N3Q}C!
z3QIScz^YijTIxkl!i|Zyp<&eHaHh=}K4U16o)uSM<HJD$B)f(4kfy|e^!Uhp{!pm_
zUuvuzl>KLs3siT79&2v<$Ztv0i18(BEt8NRr4i|S>R;7Vs2#PL>TI1A?j-iHSF$C^
zpNMPL8bG6n<n4<O4|O(E#AvY`)z9*s8g2XmPxc4N4vLe;;dD#1ok7(R9x?gfSqrUa
zqD|x_VKmY~=<9E5n{E8)og{arLL!+y*y@q0TTcjP{2Y9oxPz8YnJ6@htj8yc6#Bl_
z;nZP7s^zsb&)!>dS>}{1ipD}+xAS-vu2D#_8iYV@qohSMtpoi2v@FU(xulV3uttrD
z%Ge%eQ`7AU)=kkhL{DUX=%Bn_uH<DXnrJDvDQSv)@a#Z7qVurYL@!4ua>=+;2JM$h
z)V2deQ#KS$B7E{9!pv_F8sRe(Hkgts%2w)l^dVI{<A2WIk$MpsjJ1`E2;6sZd>SN=
z5!>2cf!Dcx)LZ*sqB5EsXyA`X2b6Vm8lOuyMc(o8f!y#isw?TXb+l(VTG$}Dw(F^!
zkMCmL@;>WL%WvdgSM|uXz;nDld#7}v@B`V+4igvn<~!zF)(ZjSi^#!9rs9BU9;ij8
zL$rNFcqBFkZbH1}-kST`nuWR%X{pyE-^iBI32xQ)-~K_0Tj;C#fBcT{`jj(~S%E11
zg{X%>Y3=X?`!=zy&>^rim>^HF{HGr*E*09^cJiOo#zvP$EKGUpYJD5x0C&mVPpm?|
z5>ACXvCpX2czG91ru$0LzF1y}54W#`aQh0o+4Kkvd3J^|I0;wb<7I^jEA^%{vZBhf
z_zD)VZ;C$?uE{;&)};=!K9(k1%u;qZB$kotAte%FzZDTmPT2ycD)?pj1Nn*UOLXB2
zb|hPOk%{4D_)PwHw1L#4v^ai><ScHyy_(%Y-VY5bDMhF9ee5s!9BCu9o_%LLUVJ|7
zYNXIJ%x;pWfxi{I#u4dEsEy)AxgY!%-DOV`_X?Ohfip&4vG=8W;Z5KiZt9Lg+j;4L
zv8(5gu}pXd4h6e|OXZ=|NB9z$mij;2%dpj=7zLzfHYD{-_^NawEmppd?;!Zn3i1x?
zD=JX3(t3!DXdVTs+O~*$$qjshLNPwzN0<7DAoLpzC~t0VoHTd9zlXlqK)Q^)RIyWd
zyqK4aP8YImPpDQ@i?lc1K~|IHIjRy)MqZ^|AxGJ!kW<79)&lXYtpnT2{>J}@dtw_v
zzA=3w8ak)2EuwkyIEo>Lqi5x9`q#lh!OVz_I8Hnxt^u!X?j7|?o?`z))!0~TY~V3s
z4<g+8@J6IDE@=ATQ@FfHq3uB+pB;zx!<$%V*IUg-1D`@EQq*@0cuRh<9W<h8TD(1%
z;dA;Y6*UN6iC)DB{5X+gFSPe0TkI;uue%3?%SX-;K1z=cf_hkKaE;L4(uw?-Hm&r&
zt*0Cng2K|0vGA<q<zbz?313Mh`L+dq8Y+kzQUz&~u!YYh$4VLE8-6bDF8vUmhrb6W
z;^URyursYXIf&Rxbc-yoUE=1D4TOdis+QdyIM{#2vrgpUW!_H0TfCgNw)wRoB5%Vs
zlf4kVEhv7K!qzpyda44MNSz|n@RRmw)_KM`WJF$o4VB;9!niXumK<!G6@3vlmc+=D
z$VT|t(gDJ4_Nn4Tc*ES{>eRJ|UYY~Yd@Ci*khnmskeD~#-JoO*TSiz#3=OuD7N&eA
z7ZWkuXJdD3ES79*Me3r7#1-~w$s_(V_{{f$U4vE+m(*;B`H%|EYT<!$Ha!Lr{e8%X
z)Kh9A@hUPz{461?j+Bc;b$c)hI;oiwZ(S+KxqHa&>>shMhd}oeP30b`v-ne-2cw~h
zco2T-iMLKgE6SY%h1|<vPtxjNLJdr*4qfF7Ff81JGy*+^!_huu9QDQjj+=t5BS$M<
zs$J1`a4iH5RL8DU8x4){W%6_JmVY=2vCYw1@(H*JS)0qW{e)@bd-0C&3J-IOgIl8>
zkHgkqXoOCU9xCY<n<A|RW6YPSJK$zJDLSCd*c7o(uv}p^d_8o6I*W{vY6lF^RAGhn
zg8<orrd?QT?goQP-6FU&j_YS*+5XrQewB4%xO}Kc65)-~2%^4E!T?VJ?&lZFGwJ%m
zSNk5~VyKc3P{{mGVo&^5;3JVPkjQQMhifiyoNviZ6<OPrNGofIZX@O^J?%B@L?|dV
zN_F`cApN<^6hi52RoO!!#rI>LrP(XD7EiOVw=GjB0+v?PPChX-%vKs6&x2SWbR?I8
zH0LtIxBcd*+LIn_tgy8eJz*+|n}$vd-w{1r+2~iHM;NDa*%Pcrcqz>wl9-l}>h|kU
zytIA$2;vr)XV%!_rFT$)6c^4Uwo_Zguf%Cf8}TnaMq-7<kwj`l__7ddYhk-5uClcw
zTgzSeO2Nj#R&Zaof;FFd5PTy4WB=tdTU6u#+#zrlYe98#zWJh(>%WaW0z%<O5v}zq
zdLB*@CJ{;UG_nFE;!fK#ya`wbA02YzpCkJ@v#~}r+qwjqgq9Psh_ymXLxs=<Vi-vh
z2lxa{iBuu-)I3Ujx;;rg5N!^wQ{wrv*&VKZ#C&`kUMn)jCP@HMi=U1+1>#YXtRMD*
zdif~(kGrK@#5t%xfp0(;{x@(I&6D4gkC10pXJn{vid+Ll$nKi`m<q08X-MI)jq8LO
z*s%l<9wctyZZrSme=#+QZP-=mRya|(CR`J>e0Bb3u?L5u8fk=uwkapiaMfrh^L}Wz
z^a|M<R+7qC83YWe;UYs>IR}G~E%G~jA@YJ~Vw?~0)Gz5F-oidfIAlL*DTm$WC88&`
z6FCGIp?{+N&5Zvcs#qepz05B-7}^c*gU*uu!+o)J(t55e-Ot=tal8q_DRzIDAW>mE
zZq$9J8e%re9oZzVG7sj)Xxdpq$xDR#rDozTb(H8s^hNT78-kNrf~!-epeK>6$iJaJ
zMAToHcoLlJTSl*N)uVfw{%3duboW8fbQR@m;mh!c6D{;<$1v01oH7N+4Ywf!I}<#s
zJqq;$J8o`Emo1!ST)Y0Ol3pwE4Wv`i4SDqpyL~~^0Hf8>Sre~q3zj&4r~JqtXfnif
za5mEah2Lnr^eX_+&!e}o*$Uq?)m`3j*?TUza_p*tGy1B|V&EO^*XcY4&lug%cu{M1
zmg=?x&%9ODd%$Uc%W<hrLmlq1<o)qBb9=IOYcOva=!y5GW4m{zO06HhwS3+^(?T#4
z{OoGzY-ZdAE%B_z1_N8@Pe3p4Ye2<>yzhJ+HH*P|s;D=_v`V}QA2ZY`cm~!dLz?}_
z#n6Yu6c=lJYs}OB)xI(9s=C<MU_+kyQDKCht;+^a`K($1tZ#njTa?z?xH`G6<+{tF
zOEkR(27B2+2Y7tST2p1i3QaS501dh;8yXe%q*rEMh-+iGRS<9R`bu>!e*<DP^A(6U
zI6OzWd8R9L+X9z)YN1U%VOtyhLel{-;^6abS^Ygbq@8q}>2C4dxCI*Znp);c-jS|6
zGpH)G94qSSTE%?O>u6QZ4SF_yGvOxvx}d8Dajw(!^i<w<*4+n6cBKVg8Lt>WsGsYX
zYtj?X`jfLaGQ~^=IKcGB-=3*$%+!~9f4HhUB;#lIUB+bC0#cR(%)OLy{z~38s*Y<B
z$Fsr-pxH4FI0x(i?cOV31IHwqBhs?+Op`Phwk%7|V%8Xv)%EiZCv*fKcwU54wKue7
zT@95ySh3H{RD`NH+c;5ryJ;a~XCC{shJT1;IK_{due$5&@^o!=JD_HP56(s2<=`Nw
zME_Vf3f$s675WFh?;GJb?t(RCoks%Qy=#4Lx&`n+wZpS3wwA(Lx5JOp0pCBq$DVi4
zB+ZVRMzM#vY;FgC55ry4enVwl1J@eukHG7=db(uaLQR>XH`-yE%L!}qvg<UY^WB$I
z=DJIuL{%`kgZaG2Zrtep=DI@j&Lk$m_*Y5$mC<K-S3$AP8J^gjC*BtZfQ=f*I<^`&
zF?s%%j#y2eH%&XR=t$D^H91ZYoav6$)l(-aB;bDi3SBE_)S-4Crc3>6!4-85JKG!g
z>XTzSB|o4a0n74Dt~Ks#I?n&p&-gy(mu!sDRdf#A(l}`qa6jp|>qY)wb+SkB9aBwU
z2)dnrB%K9(D$WQ@bfvpzZS<#v>G9g7{@nPpn%+ehgg(r2rj5s;T?YJh>Agw7_hizT
zt@~aiS&E^bxdR=eT<OND`R_DWjgY2^(*4L(x;7_iie7CV@3O0A=_{IJX_vmOXE+V%
z)177X>b2EOS<e2NW~y?9Eur5~GC0C}Ts29*Kl^U*D7s{G1L)wE-v-vZz|hOloj$2u
z=B-;a8~gy|Y4$p=8=zvFds*&W?^1_dRfXQ5{_88y$oe<M+vq8P&p17|v9Sw1+w<F<
z;C-O21Xj1~GDw<+8bH3SnJLqGH-VG7W5^BWIot?X?wpNe_@=5)`N|s3nzk1YF{GQe
zy83(T7XCFX*?h<0H~5$V%5JfU?&wdE%j+jqbn6Gf7nG|o*1Z@AFx~0T!5;a;`NgI+
zMJrVo9ahC}Kv|gLX5M5^g7cNz1F$O06jw0I^R#FU5K<DlgM4>V?ibBW_|ISD-bFV6
zDrrBec<rLiV^c8CW8Y~YQsk!HVA<@&a4o1+u{Gt6kMMO;eT_eyT;6ml>1OgMuStK@
zYfs*s=mstra*B!KU%+MOvBGDSf8<txUg~!OH%ySP2JkXxDDzY^*VxlvSkTq*uWqEa
zsT0xvEV|;oM7Pk_FoknKpWxCl6-}kO4VoRgg++uZ)72S1>uZ;LFi8jGm_PBGy|l4z
z;U8cOUCuazv1?*zILia92R>@vI7cQz8|o_c_&+&s#fF9x&g!P6hI876ZgYiU+^0|n
z(^~I7Q&Zqm{2uU$LJuXS94y#iTv(v>bk}ruIDHl1arwz~-`Y=nVUv#TEdF&KOUd;`
zYc}?et^Bo0vG1>Msb{!u0k9lQbFDJ!5_@W{`{(MT-bUU($>UUl!wOE--6-zM)K2ul
z0nL1Ip=ms5DeRNI!cElVU1u~81ABm$kkP{$`U0Y2xGbaY<saydb!2QT5BE%|VIG$9
z7ucyV5V6pD-EH}=cZKmvqRh0>XF4Vut9st6va}bCGl2od#>Hoxk98Z^EMo#QG%4h5
zpi5K_^w&(RPoIRE+LjPkQ(8elQ9rPji}j3GTnkF?)!fCiK*=Z1%!(#wK|{@p;yA;9
z%2R6=v&nhdV={S*^`_OHM$mJ$$c&}OIFhvuXf@adY?qSkf6b_%7l0ZXTWByQnwu15
zxD4vcIX=@N?F7eMBL~>Qe{xs)(zNB0QdJu~1qFZbg+(6{^u9@OV_jYCzR*R-5Eo%^
z<()T4waObSB|g*~Qciw1*bLnq(>hsgSa0^`PIP!x(>#04bo^a<u0O`OD6YP77X8Xl
zg&UbNH>Pgx4Ba~51ki)-qMK9Dl7WDMeq&}m{X7>(SAkNse_T34myJi!)22p@v(`da
z7426~R?$eM52Y>KX<VQll+r@CE%>n9N~nhW95cb67Wm>{?#*}Z1GCvSCbhSBOvqKe
z@TX^a;t6jtFfR6AQxxjpxuGW%(!aVxLHU8D@o!CJ0MfJWFGZs?Z5_L`qv`J2xm!-C
z?>nEl0e2PpDzgjx1=KDc9rL8Po8ywZua5#Jx(ag&igtkulP9Yif@Wrt7fKo8>18~U
z+tK*V&{lO#+s<<ZY?d;DF43)_(;P=UcGGKyPPwCBmJ4cY83Vpo1qU2`Q{p|x<3>pR
zT{GeF+v<4Mx$Pc5v(H!P_Uj_fDj>zokMHKIWh6aqpo!XPHM_fJCgX;9{%wuDTAO<S
z!+XvYb><r<)yetfee8&W|EhX;2RqBVTP9b|`&?i%4ARoBtds}aPPw`hwQ~u7qCd@7
z7kFmU`kHHAR3hm(Lmlr8%_{n_|D++lIJXegHiWX=eG9)Pt!ENERY4bhJNdpDtJ4fL
zIKG)CCvm#pY>N7!>x_Py;>dVyF7KXB@AG`cu!3>k6ys)1b8s)<DxB?V0$0$~({%)w
zD6G~M&9Ianj%2?*>q~4WeK#E%yzXvb%5;2DtHH5}7EhOg7=0g=I-oHS!A#GO8tI!S
zCsbjQ^R7GggN!=^y6lNjjW8}${nq}{4)M)b^Vw5?w@`Q8Bf6y~KQPL4$Po0V(P4)P
zJPRfpcj^v;c)q}la9Fih4O5&S!44)*b|v**%?Da<SfhRKcDq&1kmt5*9P`Pu5qzgk
zHvQ++8En8MLwB$v+}F8TUCzJT)q@$8{5N?CxRN1@;{CsBR8-i$dw>vkj{ZU$0EZ_V
zt7s}_dT%(P%1b#-cl6Y#eR0cLUq3^Y@J;53Z%lHIaj6eCj7+J{-d|sqJ`i~Aahm=s
zCh3um^S-Qt98bq0yH^9-pow$}V*(eZAeJSx$2&v65xA#NL7eB?woSgQBAi~MZ2%(1
z=J0#>f^Cn~2lQI&aevF~pGD{6hj|)0DtWTEUH6RB56p$sHInNY%R;Xm1A&decpsv2
zdS}qLO!MQKX>+_u9?5A?T}#<ioNPwb`6hGB2d0W{PcmQh5=itN_TE#?%dMGIJ2}ME
zGmc#`$0sCDa4&aP2TaU1$7SP#+}EbAz-fOAt<gIww#1a`)dLaL8O;Z^%}8ii!z|Aj
zV4rIp*xk6&$Iw1wS@&c&U)YH$DejSbybe{cmhR!GoIfq_)_2b_JNX}X-!R~H(kFrT
zfiAkC9$)SapqlSFc*J~w-ss!o@#W<QqVBkai@s&S-)dH)%{icpFGl0Ilr<o}E>p>Y
zwNBco`|RxLs%Cm`(E5%tyPan@PpEN(8RA?eURT(}&pJzCB{0dWW7avIuUp~Suc@O;
zaCW1!HF2I@^ip$cu(NH7+h064+3shOd3{UGOYe5L0g&m68fqDifUN<Ze^@*!{-}G5
z=|*q~G|JJ|KiA~*{xVEg*zb9p4W3tqRfcNpWagi&va~la%2xm$P%IivJ?jkhT@m`<
zqK@X9o`ZS606lQhan>l)Ep%I4?P#~YonjRD2sROyr`*xD(%b|u=(jW3ac}9Wx*Ewk
zrdEC%@3QQDMK*Avyxy?^s#EKtf1vm{<z|YhW;G})`-Dm~_KX$)p{z2D%y4?ObFFJ!
zjWY#sj@?n-MI;U|Kk)owD(mK(t-#<SS)b2*)K3PxJLVM+3%oPlrM*Rsu=D!IPKfEN
zi*cVbJWX1YpmKNv!_`lXRm_3pp~Vk`LH@hUGDEVGh;6KG={fIv1Yb;O7W>va%+{`4
zO?QlGBv925^^{{4m?L!{^E!v@T4`wLJImylY8r2Ak~A9(gG1FpT-C(eS*I>=LalW*
zJm>k18wcf$(b#nl)SoO}pgEvd$%Ievwqownw831o*EH1Vb^kTa3=Gpp4Q-q?+-)>N
zox6%M6Sspsb=T2&04_560N{1<Y9q-UR-Y=U>{`m;%WG-jqzl@;z-C~a@u{nwb{=!o
z*TcI#ISuQx<*4UTKADiE*=o1~&2;@u`Jm4OPP-<#H)Vl<pUDFBe$iXMzz~SfBUa7?
zH>J3=cY$#RckV4$N3fr7sd>$sIl80jT=0u;jQftOqSNO{Qr;>Y>WEIH+vL{*M<>5d
z*3$=)GE6b<dq6Ee%G@v?^!%gEfKDknZqPK)-5312uF!i?@eA|@FBCOOt_KWufR3&D
z5ylCCY=pgCpmKo+?ll$zq;|D6SM_72vXMM6&cqu!Cv5^>dq)Qn{G7SItCM$>tDz^s
zTuIlV@T22uN;LprS_6w2K{yxt(BMxtc$ev3CQo2y&@^zyxYPN+q~pHY>uc+kQ@_Ju
zjOu4QKA6JbJ^GkH8TJ5W48Pq$&EA;buDRYHDU}QF>K3zcp3j=83gfyWpoQNQuXP>r
zQmTPPx%qW{`z?dFolvq1h|yY91w5~7Nb3w*?`qGCqTGZzSUu(rSk*a2>vUzf_G;U?
zj~f!8J_Tot)4=5$Z|UopdU=}rI(Z9|>v$$_>+G2B+il+F2GzO6EeruV2JD)D*1KMN
z)VB;!Rr7<X#f^%xyh9Vd=Zwpl46F}M)=`ET>aFopJabI%^D8ndyvMXVQd;UPx+3>7
zb&_K}IMV%1b86e5f*e(Ss*JZhUD<gWNOBUcE&vK0_utlQ3v1KgwcEXCfYrX{z)=n6
z+TeMsZtOHWUOL}p*77(sFFY#}78DHd)=9P|=<_?da$=^XoG#wvyWzN7cwhg~(Tn+<
z(|qemaA&;E&8C?0I{^;nl&N(=N~|{Fu!+t`of+yWb$aP(Z5f9><)zE6t>TY2?btlo
z_0m+uJ;Ifj)xo<~ThsWR_Br!BNAhN=_qcLttK(;Z!TGQIrDw5g7(JTKPkw2b1<Xj8
z;m5XVmBQC{$4g%~&2vowFxGckecCt9Wz&!M4Wes!ZqZYUbnXc$KY+bZ&syaQKPA4P
zYkO`NyPQMxvUY)ksk+=TO`F6N(?bh)m^uM*NdxH9bwXQ$g?E6YO6i{c$*sK0H-hnq
zri^wjFvl>4#$CAM%cfHOh1f~p*P`<7*_vX<3+-Zs4PD@9LA!jn=_<Yts-v0$O%Hlk
z;hH$3H*aH8dWOLem!-Sz*rTbEw}3wH)an|>IUFzZ6CC%Q$?p346c+}zP*Vjd`gVqE
z4y%8bQw!b!?7D^dW1RIpukyASyoC>o+AzI6sfJ8&nCe2gPpY8fqzP7CQAPCyP8YDS
za9Z9i&(+Fbk~`@-(F(O&m}HpbALqQJ0SZes(_KY13+ct080LZ3ZG=>W;yem_)rOMd
z3(78gps=g|yrHjQqTz>A<9(s+?tM~NoX`kttoxJDa7ACGQi2q1a^uA-HH}=Obi=lh
z-gcUMnv=2pjgrUYa(O|wU??trSk%xJ2e|Yr6LYpnU?XQIe?P-%#a-Sy5B4AhLgKFY
zu_?yfiFp;k``+J$<Gm?~hn%dcmFF=H<kcv6t?!$EQ3q<vuj{Kmkx<oFtm)?{@+|Pq
z(bV!R1amb{v`z9xy1wa)Hpp5SjOec_^zZV5R|OlqixaXvQ|Nlhr?jg*KY$!Z3&%m#
zHA|eQCUnA==$aWn+fydt>c%m??W%Pt3iIj#i{pzHc^moiU2hyq>FtJUz!hMDW~}!f
zc)QSC^xX9!=40UkHEuF9Z@pjj%T0fCKNUq?i*hHXOad^&EM-Tgb!C&|UCkX^J-w=A
zrZg?cs<R;Zz9x&Fk)X=^?ua=4i$7kxQPra;kUY*kLpkl;-g=1MprnH*_|pr2Cl1zi
zN!G+{s1xUzo84_=9*tJ~qnW4A(B7ltbTNgg1uElcr;~O#sv7i$$+{aUd)NQfHv~Q;
zT8b9Zot=tf*4<lE5TETE<v8Fuv26fzsQ6bLscU3PGuVx%d_oFQbeF#1IP4p)?xx-9
zXaO8{3}eRU`x6P*6{lTyy-v$~SK&B%x{+Ids94+<jNfC>0DY2p`dvy3(*Y%{UteDo
zqTvcmYs4g7!6%66p{I0`=e=nv)>YhPUN6Q0^;taJBrq7Q6smw!5f`Czq#?*fQ!9KG
z(G9&Kq$B-36~J>$l7w(&r9bpq(BtE9FmbNyrnREdDc!Bf)Q73{Arn!DI7iRQ{$sBV
zU-cEkvssgSh1rTe@RdovhUMgMVI|><cv*E3Z9*;+sw3b1o7s0{o5HT46M=2?KK!8E
z5P3<E=2=3L=MXk9Faqqr&r!|u*W3^v+MxL*RL4642LBD@1WOQEXs)$6Fe31SJ736<
z712oQ@6hW|+q@YDmYv3x7e?blq0`6{Yd3B$F$x+FopE*w#-lxnh1fT|tZ<m`<9&qv
zWcGrUSUYwRk9dusj2__52@&#J@T}5jy2u>zYo%8W_&{M^9cn;ez5yjw!dP~<xC0Kb
z?~qUMZZ;R)g2W2LiH3X+wx_w8xPmx>L>!fb8?Ibm`Hbd*-l+$*q1VDuB7saHjtA>O
z6}>He35sLov)G=lr<hyEbJcc?3SP7PHKb_}vd9pIP5481CA<|^*|Inx$$1<yQBh-a
z?l2t7&R{apwNQIvG8-Gn_Wy?dB<-?HC<wyukW;}wzRtn{zGdKR`ZKNti`5YEQfCb@
z#+L`)(|!r*#oO2*fyd2?<Ku;Q3^A0w1lH!(qIdY4#ruLP7(>r;4J~c)F?bBK75mpR
zD3C|C7S@};3Nsz|@U3E9qO!V&|5@NcFay521;yWn*0K{F7r0?Wg0Mi}Rs4WN;P+TD
zya!(?s4Q@r;_MY4>APEA;{rSoyBACYXN6?m890udbUus^k&W@^!We63c{V)CUqdWr
z&Ps$d6-M}pu(Hz7ON`pQ4-W`!G%N5%QW<w`qOvfQyCZSpYCHs2FHS@6;%V?<uAgPK
z=LFa|IG7J$9pSnB1;aNsRf^LMp`E(#%q>Q(-GbH?8GN?>n;|*)8hi`ca1NSMgamdg
zU8aA{)72)s-tw1O$+r;?s{i4RTK1Ax*z2A|c!SD_#?ZgTd7fE}1=))};LFfv%lVok
zalzYADOMjYoNc~_H#7XS+zxqoD|#785IzgTL)}m(+%c2{^!DFiE{iT*b|BOISAfj>
z(W-D3-jy31qQYmux9~$`vGIp+8`u>79=Hc@<-W0M*n%%KuJo9VS)x@u$~6Fo!%a%9
zcrJSztzS4*^QWL+c(L#w)6jnrAH<r;)$kKB%#CG<*!PHzorW02zW9CQfw$N^3&Mb@
zXcJ}aZ%vv>pzh(`RZNJp$hE-(Vt)ojmWL(;_W()8zou90U-m59+BcCeWl_8(QN=Z!
z?t)iDA3&AKOUe>bp)Er>cy{0+x<CVmwsDDEMZk!cC$}pc%}%fu8{_FA%Hf;%Ioc)v
zV;K>81-Fr!i64bip7n;U{A&Lu{+866eXC?z*TeBr6=-Er3nDJ`L7|weY(v|9!Bi)a
z7=mR&>+zb@ddpM1Z>XOohsbg@AYXWYA#bH~;Ch_$8=_OR!-JcUuA%>!x6D=L5Yk#&
zWIm1*;u++2yj-9%_r@>+-z<)R?^>RfeA%8G7%FY$%WxmePsE=7e4&-)GaM5rXS!kS
zNSp$5{1>-=Fn0*;6J`ZkaPz=sPz-C&vr%`<Ly%F%PuzE3wm8i(o6T`h;u~{wFjcYp
zV?c-@rIGY8_O3|s?>HcLDzEl|!Vbq|t~T()b%$8s|H3zf8-TU2CxV9CiZ;i(2`5zL
zx#OmmN;+yJnF{R{rYjy*+0Q53hcW`Ok)H4nv$J9yt}8CXMK%7gowC>Qdzs<P7`SQJ
zWaZfUSX(eG42)%n>mdy=7>2Fu*+6hZsFvcJ&4$}^GYA9qDA>(%CulU$d@r~uzuvqg
zVpKQ7_F6RJIjkAmky;|zAl%xExWUd)(o0V9jZwDULi#{y!cTT3yi$*#lN_@E2s&Y2
zRyx+n6=BA?$Uwy)Tn1fBj|qG<7r?vpb|{0|Wu6>rCTdy5z=sxDvV~bt4c$t-Q?05y
z)|&@l4J_^O6-pL%h)|uJ;8=v74PIm`GK=s5b+`Jza{~?8%CBmPo<$Z1E5Wm&4s@QI
z^ayx!y0E0Fe<!#<;1?QVf3VBJapuI}WYhmhy>v!Sf^#?unBnam9?tfH>!3>@0vj%V
zSNu$AK3ya$BSHKhhOl)4tneVd8aI%Wq_5^{j@s-R1d@;GrU1WCZQ=ObW?Xw4#%2cE
z;%njEVgWnYTaRjvd;HGeQXvsj3#V-rO!afOTgJ=7JU{&f@ExY6FV(U+h=H@+C%_`)
zJJ?iZFb{%i`eh*n_|5F+KOuGb@`2d2{Xx+%iS5Eh`SQ4vI49nR%lkVD2LceYRhi{b
zpFJnAfq85$Byz|2Gbxv(U2uZXn`~E59Uo+AEl)%5!F$YU*z1CLW_5TQdkvZ(Jap9c
z-;Ou4nd~=wMs<1l5YH{-s=9OVHP^)wB|CxxrRVr6CGt9%JAv;K+gV?Q>-+l=I{$Zm
zfo>URvfh`z36H@TygPFz^emWSvazJzDiktGKM<}Z)<BtnhC4vGz1rYnf3jn&?8ZjI
zT@j1I5(mtUZTrnMbKJPqTPlS_n(>?7p{KdWaCPP=*f%)JmFD>u?82L?Xs8y^mex3K
z7l+8aMN8C88;^T%+)+k|I!;2f$bX<VY&WVIbqO6|{%mZ-X~=%cU9pbr75qdyjSaVT
zWUKiqfm_%xun4;u2XQ^HH&6`z7<qu#3XFpqi*`1Jnjzi|jj<fX$^{HUOX({nA=Tw!
zQjk|0TkC#cE3mQTTmCNji>N?UO+74I4CjKI1W^27$%}qRk0Q0js%R#h%XZ{@gR25Y
zU_82!nTTf~>qHxJ7CEUpWtjj6#D!c{Ou{ju2+0&~GVSq+%AAZ3rizYckYI5bn*tNr
z;d~-etfc!Y3N4xMfqcJCy)bPJXs6BuXNS^}`)KO+58^WB2=^D!z#edv_k=gja5Li~
ze9hSh{NP+qWivOkHwkxbdT5t45>F1Fv_8ViP+D%5eh}Co#l+qVb;oYNtfifIFnI^-
zYu<v+$veTeG;b4&98Ch#Yh?1PIfic|-=+H!cU%ebP0J;JeQU^=Mzmum@B#3tHWqm(
z=-GO3AT-4E9C?QJhr4>4u_Mt`bb;zlAs%u9Q!xs<iEg)-U^NU$s5bBc|5uj$ABE0D
z2iOQdQFg;SLf6oi@NYteoyGrWIHs&34LcOc!yf52S;h$CgoivK^@hvB^*yzq#!OP^
zlJ6<A%5~MUJ6HyRkTRsNcr$gz^U<)`S02B_ZiYjFih>SK<L|lS!v)|5{z<3|8V{Pe
zOTvTTY`i<^gC4LG0$cqS7%-0)<{*9b1K3NzGWi5>LwG0`8DV^|saf!{;U(P0Hy^<)
zk8%qTi0=fCrWyPfk?8-46np=}wERnPVhPJ{B^H6ZSB#SG@?V5J^bA)O{GR;SP+-c%
zooIoeLL6Zgdq-Q3+l@~}mmxd6-cWz$3=x&)6MGpKa*EBwR&#xWmvN+sh6()>!Yp?K
zItr(`y{s93z)Xcd2R+DY0<)OB+XNK-jb9MPf;aJZp$+(b=|AZaR^NY^<FLp2<4hv!
z!haJ*7Nugvzre5SUBUvh4Uc82hF+2NL)($q;9oXNoF&1)F>{tYMy`mpH(%$Tl<vm=
zm}~G^@=OBeDJc$i<1wmfwggK<yqaYNSc%W&4j|8X%HNu5@2e7w)x0ATu?_Hhc2aP@
zdmT9+KJGMx&U5SdZ@5h=BfSZjp*oI#+4})APV1(^2gp1+5nafygi1+uq)cePGQA_@
z+bN7>e~FF|BaIhQF(>wjy8%C7|3f#!AoxMpXCJM3rCOpP$#qmF)!BRno0r5U9U-G(
zkRR%=R(y==ge9?VakO&ZXNa5p+tAw50c{3*J3B);`}QXKu+wz;Ol#|N|1%#XG%M+i
z=Q7>7<MatM4x1<4;Zme}zL|yZ$t8x?a8=1hR12K~<H!@qwfU|^o1n^}F`*LZgdsk%
zg6}Duz=x8)&;optyB*QeyUid5<ACCAyniTt5U;-Lmw7+7C-OjkOHO2FnCgUUo5P{F
zl#YIh=d8a&2e=F9H{>BWq{iXmI3AOH)J6YCeHG>`H>_erB+OliZA8BRCp=bI7TATW
zBt6>10Kp_<hAM~6aGUhkMfpN?<5}V)+uPw5J8+AXc@i_Bex$YTQS~c+g|Nc8R5*%d
zASzgfkNck&4GnCt^q@-v1bqYRZ8@r1i5!tv>uUwaIR~@FirM-C(~P|xse|SO+<YBh
z|HK7$FZI<jlz(DwVrsxpfzKX<{1K>Q_32LHW4L<G8NvVX@~NGKWI2`jm8{1hz6D+u
z>dIfl$HH#}KcSP}3QxIbVIsH!j~DI-_YhjF8GwjAwMCH6pC1{I4r6DP{A;_$z-AG7
zfvz?m^k0K>;YRq0!p`nFc&xmJI4U+_D>Dg((@0aSTsTfy1=~zr#R7Cj`JdERVLxd?
zcY8;0bJ4oOWqe5JUumXzz#<4M{2$r+wepGDc&G4gvK_RRoMx#SY7Th>8^;PSMLW6(
znUoXcDw+lYFG58+!7?)Ux>PMA2<=0Lq4nS!{-)v;Vg?@2UDe-!G3iwT=_vvBc<w=)
z1q0Rq{>4rS^~l{%G&5VBBaB<3oT)k2!Fn@Tg%ZtUkzPzULq}<WblfzP4syGM6GHX$
zb>eOIH(Hq+?aO8Jq%v_IusCXRCAHHMj9QXaneyx8#li9DCBs;IEmvnpW4RZ8x@Hdd
zqUg1|62arggYA*JcsqVAKa{9$vWCtuqV$0zgaN`pxoU7sahxL>$|o8j@62NZH<bIb
ziET67N<fuN<X&GUQ%=|;za`?))&8R#%Wa6nVfASXzcExDD926a$BN_FsxW}G#Cpi*
z@JrHi_!{<>UxhWHx(n@sJCtc_^NDM8XstRSA2L`Pl9&7$&RDoM%BLv3Lq@_jB6{Qm
zdPMOl1N>8CPhqn3&^682J3L#eOjP1t8#-FDka65Wpev9by2SjXx@vX78~i3>s89`^
z9$bzd6{Y}3;c`q8J^{voBm8|h4q5MQB!ET}8xm&GO_WMad)HpBY3MWZ)j5Kz?YRot
z{I@ZI{%Y)KHlcRl8WU?dp41=u0ijH$848Xmu%I63fZGnA;qL`l^%<@rJT$aMEMgf`
z`Ru=ZKO_cnArgNAVna^wuz^O>VZphLz38eH_`}cS`@xfiBXAE(u5KWn9NNjnTZ)BY
z#4av{Z>Lq^)A7|co3N?O0j8dLLdsI)GS}7cP$&d4^nJNKa4&ue|HjnU+|$yX=&vCF
zD;o5gt!_tW|3Kvoa5$7|S%ruFO?7$D82Ao)iLa%Y`b&LdfT#Q&AS1v4Wh`6yem*__
zg&!rjksHP`=3AataDx6OQaACiPsPt<z0N<DYjmPW`7z-Sq9eb;ua$peXK<)8@nIH!
z9DIonat{S+D|bUZbc2!-Z*Q@Lrm)|*8hjFW)_2X9!PZNdjw}I3a6v8&qd^Am0&Ry+
zTC|39p>BbV?EXN6`@>q5Po87<7;7U#+1l_+HZA;`T?Tdm@AxzMYPuPCT`tR07hUVE
z?E8=7`FX`_a59)^><&NhyyX~B5_d^-;0p2<du-Y6(i+~dMvF&DWcD-P=4<=&LqRUX
z`qZz%452dSr%I83cHjg0%n(z46#6hY!zaR{JwJV`oR6*L@BylKLK${2eiU1Z%n4O;
zPx4F&n84@eCqgGvtAq-*I|7iq6U@7gX?)BV|2|!|@wp|P1NlpMCu*Q!ns+k))ZEfS
z`F8Rv{Z|oFC|@@Pdd&?5Uz^Lht{OFAk$carHCdF}P;p=+)DBre%)(R9gUQwSjYyuc
zwlyoXKd{<&jtd~KnLF4F^KHkvU~9u0=B0SS$+4G1n>h`;$iE=B3|`s4lu5uYqIRj>
z>LW}gq(jO~=@_pOt_c4jv*FKtQfQ@E&IGu+1})rg{Yu<qB3)6m88R46v+{xt{m3;z
zn|Qu65YmdT>068UM$QIk$m8pVPjvMm@W48_7`<qy1M3VcjfxyfxNGWx|G^%cgGJRO
zlt<8AevDZy{q)@REd+ZP)?wY;9AtubzTtj21$+Xw7M2!_RB8-M;8L!EXfjs}1lXO*
zyuIVta;!g8llw2wLhHlN1&;7#`9#w$co%vbo@ksHyrMhCeuDHyFIY`jtxTRVnF028
z#fSwr=%gvyPwb<V-R^A`OsvWKnF)z4LOXE>+Z0*@uT^sRb&!pG96@5cy^RA)V9=4F
z`{r559Rs(RTf&=z<@Gz59ziI$&@=`6Ybg=?DptzInq`z%m}yo1LvkKl=-(F+q4ECF
z*eb%x?P9(}pW0T3)42*@jxEDEM;mLihielKtP#7*R*xdd5O;yV$xfwnw#QRL<hp6c
z!iPOpvVyWtW}(A`uvAAp!#2}8q^4}V<4kx6UO)XgvVpr{_7REXu%s1MkBF0DTUFt{
z^<?l79P67;O|_TDpHek(Tcjef$l8`V#phc0SO?i#!9C=B+c;Z!ZZWmU`(Lzd2r32G
zuS92Q0=vRCg?d^i#=K8A2fK%7Tc-2z#C&!gtcSHyfvs~%_W!@Xa9G-NqBcJp(OOrA
zdzD<qf0)mvy)d4ZE*N{6?xuH3OUmdcZ>L6vUm{CAuf$NG1HKV%N3DrC5vHVq-HTO5
z62zXI0EhTF>-Dm0hfgL#(<mPn8*D<@!_z|@g;Qy3<Ogz{NY8K`U$XQMS;L+zZIHwM
zPRbrs&eE95kgr<9ws$-4lOH0dsSo~?jP-_W>IsvG)4@S$-$HAJa`+lB!M+$T8$1F-
zcoEzRZX}#E>uq<~ufpy0n&B_D85o3(6GzJng`Pe&lZQ8x|43WHC#Bo+JJA|xfelZ+
z3iEJR*w3BT*CV3dO86%3rF4?KL`i91;7BQio)0(Us8lnbr{qO1fNRku_*M3pP@7s3
zu!SE*SB1xWt5{c;;+~r1I-Y}!%q4;{udvjx4yWcJ`f!6?r(8JRrsT4{v7vEk1?i@U
zVEgzK>8bFAN|2`!i@9EbYL?IPcl06hh?*9CV6F)rjSR;pqdSSrJsD}Oy-<maZB=T>
z0J};YTe8OAo!nW{HnKUC!?$N{rN(9$h+Fs`bCBFAPYw<c<Kh3w8%lN1W9E=`lH8x2
z1rOfQD8Q612*+|$jnDb3{<9Is_7}ic^mnMLXt9osw2JKIhZxV}{|F^^hB%3wk=h3R
zVC~cwiN8v+<V(?^P;F)Bb_u)7Y)Pf9ipdGu?{8+y=RZ^BBQU-tVwe60H%WcIy(!x+
zh=i769J56|07b-&R5cr`?k0zn=|-ak4LpkM$xadu3!OF9;an~zIGi%UYvJlZd0}vP
zh;-U=z*_;C=)U1<i4Ttq5sr}aBYARl@*v)n>ni^%jSaUbX%rs8pyo%mc=#jL%W{ph
z;(gRJ<Gayz@d5V`){9j|DKsMxOLQThV=Jid;xaT_c9&$>8g7?uE9l3tLtz4IMPJKJ
zp{MM3{EAH4%B2+xPl9toFX0OCQO9J7$7d<CgfqyM(VJi-a-67?RuOw{jz#nFa3n+-
z!#Tn4*8RnILu(`*`xIY-UnVMxvyr8+9-nSOh{wzqcr1S>9mTX%jx?D6B{IUG&~b8=
zd@edwb_$i#0A=3l?9#o$*6;>$4Dn7bFAfr$+q~8osXcpyJnuUw4G2|B+6NCL8iZ|W
zcj088+WMYqWa}0!ByL2#lF!y(nIO25e~lX>_3ZC0!?<@LSII-EW$MMW)>H&tEdi;A
z?D4`wB7J+Z&^)}<v_b6e-yu3-4cbzC5>TOI3jVsr1Y9(4Sp@BrY@zg$Vzxr`mNnHr
z0zHeb2%f+vrftTqr#+&65*hHI@Da48Y|y;I6Yy`ub*KZ=Qtl>|;GOYXVUS&oCP~fx
zGd*G|Moz@@@y*JN!u*swJSe{i$H0EX16~$;OW(vp0?TJ0H}N5=M(AZKo4>`BB@S8U
zFl*4p(Uq=~!b_nPoM6kgb%FXY9k^EEQt7gM6sSW?@inwAk(2l?#Atp2^ja*Juv5Ap
z{Ffk!2?}HN#rrqp6z9PjVv}Yl+?Cu!)Q9sd6qF&>=gT8aOI~^3M^1Y)m(Q}8@VW3g
zX@KsQuv<uWAC5r0A=;npKvlOQ^1ajy(Imjw18_;no5G61K(VXoRtP6CdpF^Ahz531
zccYE?CFm0UlG1GNDq@DHE2<4kk$utI;kjhS_Q&#CCZQAoYEeJK_ryNbXTm7Bq{a4+
z$}HHM@C5%P=!Lu;S)<>J$A}S9Yx${+O&eu9qU6rGz)<`xHUqDMGogkB{n<eZ>2DEd
z!rIUS;uPUFW}(?qWudC@GA$1|Ph{e8I0ApQmJ;tgL5(+Z32vP_*L;Jjga1m6VP;@1
zVlm_tPTDdfsdCx0+wyTU&M?v+a|+S{A8VaV><x*Q!B)|b7P^AWL8ii0X=D9Yb`N>S
zT-U!QHOs#+(A8Qm)XUZi?}c9%uUVG|>nhdAY>PED#oZ$`7@5t!5qjG8P~WADNKZi}
zWm^Ui>xhZbdi-AO8P_+eLzW`9_*6|B^0z4u9>+k|QpO}#GEWOzs3W$5tsCv%qiS+#
zC?D^iTu&TfN#o9Lokd({Yj82bTJvSxFFvUERceEe!Vf3zg&GJ`#6|eoz;Rm_yk5px
z;-;xf#uPjir^zA-k}ptItu0&|(-u)Uu@!$O4zb6}J`}TkR=mVD6gT4pv5o(t{URt+
zqZQI*Jda9~ZRI`pr6t%K@{zZ-P#vuf&&G$?)l6BYIXKVOxTIa$BmWO`6|$c^P^iUD
zrz+#EY>k3D6h^5?-l-b^)g=m;r<P}EL#8v)5w}thG|PHhh=ymA2hn4;n|PR*=RHn2
z^aqj7Bw_m=NkMuS?8Y0h{be;fQJS!>4^=(w8gf*by?IRHJ-r~^j%Dl>*$Va5oG$rS
z%9I<4&7uz-^9$0^Bs#UEo7kpgaAY>4QchHj_$t69vX}hIn2KdvQbCnMjSUxb;IAb=
zqUG)D@jo(z%_X-HS)@*RWZ$!GwhgI=%m0!+<bG08xFB_YWGDNA)WQ+=VWd)M5m^|i
zPB*j6jV9Y#3&oM+p=n54TaVJM%8S@-ssmSE?ie0bG9MvuE&es?fJd8d$Onad*b~@K
zebKwbS*0db4dR11Lw*l*D@kHMm%OL;Dc^W?WQmY1&bA&Do(X44DxoWdnfxK^IiXNO
z?8{T(NCaOY?-6#b`z*dD26}#xM!b96XZ##kF8!oodPpK(QCFywRy9{6b)?`kx$*rb
zw-DamCNL3Mhg*@R(rBuNxJiml>xnnPNTeN<r%XtMEG<D3`6PV@ejp8(YuFm5HVStp
zX<?zfpIcLMK*%n-hBX%YM;2RudiL{2qW{;>ng7Rh^>JJ)FH{k#6_LeEa%bi)=iFr`
z(Ml>&JcuGlDQ#3O)hK-^A(EC_LxoUMWDzryS?+S~z30vh2`Xx7HI_CJN}rZ$Yp>*q
zrKQn4p8F4+pU!!m@9X>he%|js@UB#?tf%||kuOacWjYZf_`2@FdC3=?_5A;|pVWKW
zAv*;-kufRb@lw-_o8|t1l0&;p`(t$I1$L4r!x^Q?>S#5a_Xif3y{S=THe9XFFt&pI
z;yUqNCB<?Jcd(O)Y;S*}%IUH!5$~d#@PwEdY*4Pc#)*^Y4n@N~15>NUs7Dg6L=R_d
z^3RN6SN(gP)qFF|H9j&jw7GJQK8&n_(8x@R)0V11n1X&r9_;~~j9Y~gxe0EvWz!d}
z8+6lBNk5U(sp;4a4!i28vc0lB8V{1JfLABs3&K-40hMU+Xea*FIZzu7@5`y`Ibkbb
z<J(GShC<wBnThP?bNH7v3s~lORXc0j?P-V|ychAi#Eyh&^(2=^H<6dMd@$NLr4`F#
z^*h`Rt&8zUtBgHOt`&Fcca_KRI6seDog#n^`D5-4e*u38^`>U}(p*zjITWK!<DW};
zLUlJ&=WnC7@?=jlUF6>))3Oe7L5@I(0-wysYLs4!V{nm{C|7`AsyZ_(Xhs?*l4z4N
z*0(bBvbdE0hV%%yIxTjY8UwGW=TV*TBO6ieopx)B>e3uT$gg2<@vCsTS{OfYz+1il
zQ19V1s?hO(ZbM|vJZ>J|gC|>uqL-9yGAV$4EOjcpQJRd9^Lbx9o36RTWz|tBON<&~
zduU%(zAataPZq0XX+2Pra6|oC8>23B&Skg5Cvijc=QKUmofO3nGXA34aFlosw1;Mi
zqphtx=fk)~TBD40Zov_%i<V=($u1*`+2Qd;aHsYq+Z}i2t6W#$I_i+K1#fi?_Fe~t
z?$&|bn8orfak<dW9Jglsmm7*}m~@i+EIfiEwKZrHL@C|$BVY`#sw#++=Zez>$xQbb
zn0L09Wg&D4(}J7rnd)^;GxMre{1^CK&P0E6Uy20WLN~!d^0j@Xw)yHMa6h~`Y@r+Y
zCXl6VWpZ(<{gZ>+O%_3`-tIXXkjZn#d(Le13c1b|<8!DJ;S~ydR&$aMn|552Cdlce
zz?H!-$Vf0tdCK-hrv}@Fa&ZJ`(huS#dMrOu|3jJWISm~QTPk@^DB~hYjMe0#^mc?y
zR><=yH^6>1zor)w=q}u!*z}hReRvs+Px0a$Jk(l_1$H9!F&e6`$AY86Zt)!TTS2kd
zPcr`k!Wmd;*?|9J{7yQxSzJ5+l-n#_0jE+b?N+5*)vIi-7|lOqyE?zoEA(6DNpai?
z@D93eyeKsr>B+@`vEC@h3FlQ@5Jr1{!37EdTd@Gk=@){vS{Ld^Qg@uO{{bB2a{ecq
zUth$>$UNQGSjhcGUly~CZz4Zvx#~HUCRO!$cV0>&a~bx)AG#(I5wIAY5F$pj90dz)
z&#FCa)6K5mlVG=q9DJ2tt7pS9RPWdn?nBn&i01VF3q=_TCL_R+n#t`zE7|k>Z~Azl
z$+^T;*SEoQHnGsCPifbBhT_?nI8vSt68Tq<sjEOo15?lyyu43AxKYYep~)Ai2#q#I
z4lLphv3CTKD2JbevqDd>fU5QFaP6RPIUn-<^!a*mRVMS)WF@8Z8LQ{G-q~5=+66nz
zH@(oz+aIuvjQ$ok+7rV~Ol9q7LY&D9$#rKN8_ddlH9D*`!f(|N@pa`z<96r~zux$q
zn&7%?__>?1kNzj3@NL9i>RGhF(jYBRx<*)TqEPf%eB=|=izTX79wjUAHL@>0sI|lm
z7som)%w3gZ#znl4iQy<u5B!wAN)9x3iSOz<at3~6L~v3x4Jy3mJc`!xE9_^I;sf7_
zmx4**7m1DN&Am<Ta(zzi2)+m$A!CFR{XqDd+iQ7@H%qPZYw-+T3YMa~_IkwHhT+?C
z559!{LJ?s%*iV0&^a1QJdeI%>*-|N$$aH60nYV5G-~r>`%8|%T)h7JcXo;_=8K#6i
zLh_?Q?JrFuzcVv}$8iZ~bG{$y@6bV<Jy*Sq#(-orExwFtB`=91@OiC}zZbAb?fN4$
zOPZxb>x;EB<W9?9va|meR1-S@ZHv8ET_;-9)zTFGuoMfssR}<Ct$=O%T=uL(SHA)~
zz-4j{IgWjkxo`S$o2d^$lW!t-n0&kcu;54<Gop9|WvVl%>cB1Z7JH39Bu=(<poRf6
ztOdJIQiDMj)2cR+m;7ba4a)(?ndHTCsyG%8kw<RstPQg*F=s}3KF)fg_l&?~Z3}x?
z`axXJ)6CL90{91<6AF9;IUwM)C!*~Z2dKrBo^i3|yGo@3@PJ&fYngwJ+Er?k173}d
z&=p(|43xT9(tOcMS6m4{O3KhO2-0_2+;5-c2oV|F60_H`7+11!bgAMYCK&6fzv&U^
zJ=1{u*mo0?p^voqP%3{>zDhg<4^_EpZgr8r0X-n@)9=&6gI@(}<R*Ux`;521_Pf>|
zjQ0MjZPFXf8CR-fh4#LCw*7*VMm?Kwir+)tR(<NrAjf}<HH-~(j=5>IpDrRZ>7no<
zTp7p*JLqRbFZkH8RQ{W{6U}<EcGWu7`XHV_F<eP<OJIn8pIKBrF@f_Jfkn1YlTUje
zqj$~PeTsV%tIOkUt#Tu|0r%9K#hpA-=c8L<zI4DC<E>Nb;dFTwHGm3B=Rp;`$@P-Y
zfbZ2c`W3u|agcWsjyqUsoo`UT72s1(H;AbflwV1;<k=3u-ReT32bks$%d13p%}i|A
zE^3SHBZ4(_K1;|hrfcAw-W@asj@gdzbA3000kjTugzM#Ex@XlPsh4jMqLOwIF7+yr
z?U|1{qe90NU^vo1rI<xtV+^yNx#Ml1TGTG+v69NHsZO;w4qUh~-IH5$kjk<?P-0k7
zf?`N}j`|bzWw0{V_nZ0R#+ZWV-`QF|horgl@k8MEF+-DVe7$;s674;`ok<tptV5W`
zZh$!OkWXW~aG!Du$y&)!ze|o5SIHt-zq?JogPI*}Vvx0SkJwK7OfH|U??0SQ=P!Xj
zwd>40=5SRrC?Z<inOc>UgU4uJ_P@)&Aw^lYyN}81&~c(v*vZaRa<tdveB%!H$nnTG
zSp3P;DUeC5rOcgn%L!E9w}*I}uCt#>R+yfZ*?X#>mmRg|6}H81P3*rp9@p8<QtcHB
z+%?uo<m;9i`UlaDBqRritJ|3pFcUgMjbu8^z=g^-x`3XlY};GSWVu_IgSu0h&;Bfx
o;w=_8H37~=PG$;m%=Xyx56(l&jM=D_7zAI38^r<Y2Wpb-|B7b6umAu6

literal 0
HcmV?d00001

diff --git a/wasted/disco_test/1.npy b/wasted/disco_test/1.npy
new file mode 100644
index 0000000000000000000000000000000000000000..b568703fc8770f8b39c90320b60ba0435a225e9d
GIT binary patch
literal 35968
zcmbSy^?TG<`*+bTQncRjiOWo8IxQ4;x5eGvp}70vQYci$CoVG?X>k^Z#ocvb*$0Qk
z{q6T(c=KD%b)Do~NzQ$LEL*$u>fEhgxwu(z^Wl-Baz{*u8yMh5qwO$ZfJaZ6K4bc@
z$wQ`0A35s(#@h^=m^-R$Ja_D{siVry5gTEsM;Po`*%V3`78?FPj;dLf5qu>zo*0U7
z;Hh9t*cE9RY-pJQu8xpIF={iPrgn%`ofW9-;bg}RITeml$^3pL+15tILt1-%f0o+B
zc9(jiexnwl(<2ij=GZst5z&N*CsNcl<T4zi4v|zzlh~b-ZulRe42PM@wj2wbvkz8R
zl)SVY&Fmb}h2N0@bz`ht*c3WsA*0jHcA~(3S&X;l${lSp)dEWg`I+@B`V47i*%AF<
zuOaNWYdMfOOIQ%fmM_~Om6}BzHr}w0lPOy#YbyCoH1oeKG25XmP(DLCvJ6Uo<RMr!
z`%&l}wOGW7K9Mck+KYkEfk=l~LvDo7N|n@`F(*APIwb}xS0kyJ-3ybYDMGbuRJl;H
z$}(8J&X=Met@-3O>wkKMNVE^b)<+x1qFHl=RzW|rR=s2~)K_h9sY_Bl(^M?Zk{3D@
zT1TA^w#%Y{JF&aSoag~$nZ-f@nL~qZ>g8C6qtlT2GFG1p?<amxi-Vf1r?Kbc$yiNq
z&)5vS9+hP5nE8aH7)Fkl&vV7Ly1bjn<evwp3CmPp<VmnW2(_dVoPDKqP&PJ5EW2|%
zVV$K81W-BFPZ2!Q(l)|SC-XTxG_=%r*#=u)`S#R5V&8AQ7rU*Viry!!Y_G6`?$7_B
z3}I(tzrvk(M7lsG`<Dm#^i9QkK?6VAaWZrPTNoLS?zGlI3PN9{W8$OC-sngwUPy8r
zRO?3cCPT?yWs1btE)JzypV@tp=~>sv9`LEGJF&)0BP)zlu-WaU#Hip5#Hh}Q{w1v1
zeZk(*?g>wHR3qkxvZyCEx9u$+QepC{<E3q8)@H;Yr3cS@w_;W7O+sstXXFj4X|NkL
z$l69sHOuOL%Sv@8G;G&@iY{WaV&Q)*xzUQ@oLwuGEz!=w4(eF&n{}r8kKN2X2p0l>
zqPsKStM`RCsu!lW{+?wF?GF0+7Q_i#v8b_<S=YQ>!{wCPL{&Dvs6N&=5)SFCThWV3
zL%5xs;~pM)Kn);r3^}&S;sY*$N|V}WE~lEAQqZv_J+l6@53o^6y7~%0%@ZIcRwMQ~
zvmPl+1F-#QB;>WujJAdJkyqg(6hy^^>|jfDf!(dfBSd5$beoukR$(tfD>7#>4*s{;
zKYU4S6kmhtPK;;fQomWV<Qo>ZZokmX*2*>sH4BKc8qq_8s43QXvZMWM$hhn6u8*<V
z_F=@6@GLbsJQbQ?+l3Y?HpXRX5c;32W$5TsHCbx1cAxu{{FeB-vlU5|wljPb4@Hbt
zCBlWPrww9LVwJEO!5DeJ{$V=JUft0Ktt%$Vp!%Dw6;;DAiR&f&f|`VfgbrkOq%y5m
zHWq2Yit-|R*zrBG*|FF5HRK50sTZPd@~3lp@l_p%c%p1gYi$!^P36>J<KX|$)9hPx
zw>=Miqt>?1A)=NQwjP$%wuo<yEy4aH<+R*q?<*S|_A-g#r|8q{xA<JSwp=CDC3*`P
zuWYbPa+gZAz<!Q@!?DQf=muK{A}wq&-lsYu7lLohMyYPtwKF3TBp<S;$#3c=^{!o{
zbm7|a8&O3j*GuD1CzZqgvn66zLfJM!Lg?O>R>Z?ttMG%c9BW5Sbu0iEhMQ8U;wdto
zj>AN;6|f~XnP_jj8?LbRk)=wcsainYCryYzb*6S<ZtAVAsodW|nCIhN(jLG=qZf)-
zhpWe4DAVi;wVk+254Jp(PZL>=NtW+YPq-*HgFA`s^|iHU$lq)ZYMgqCiYG^heU@sr
z9(tG6k9^vuW^NbT1v?1WsbzA!eJjykc_ydERz+?Uyo^rHd;&u3zQ|U2x;??ZjcB@I
z7LjD%WNQ_%IT*)DvQxIllC8D~%i>D(W^!U|gXv`S3Q^6r$~pu~bUadA%08RHx?C~J
zE5JP4#|T8OEIlN)R#WBWsmFqAvOfZ06}8_0I@%`7pNXFG0P>i1Nu(e6MBHRu1sHc_
z*zb~uq$2Xu&bhkz*&{3i#hcb9LTAe3?ZDsTQ&c5b2?{CvmKQ8$%OiK&zKNjHAw?v{
zs&At&m8Y30$U%ENKa1QGIZg$Us!A-=Ywr`hZq6oCoke?Y$OgdgtIoV4O!n)sY3Q1;
zpHi$pvew0z=)-6Qb)fXd8j(k4bs-TfSFM?)4YL+YNd@<s&`&Z4<1BHu3*_QZ1$;$t
z6rT}Z4IK&BCp6TeT?^|sQEr9%Sqfus<w1^g@<L`g;ji%H$e+>o*%{FhxGVaMuMj*M
z8fBl7*_aw<8y{?lB*t>Z2jpQ%qHafQ$_Z+^HBsIlDY8982U174?bKSKM^*#Ax7x_z
zl<#E*?Y*<kMlMHas;z8RPG(;-Op6Rv9vg3lN5Ng_+?;p#eR8^`zD?r?qy%L~__k2n
zywRG3e=+1!3mp$|cV@2gtiIoBpiYzu;i^KWa=)NGQ4<TOBdB_@I8mYg4)szJ<xl<#
z=4MoH#9{x)t)U)@y_J=U6zZ7V7Md5E9)6_sa}84J2CwOR@`DX=xKjVP`YW>sRcZ&s
zzb)s9(9VfU>kv)Mw<%Hwc{~sh+N+Dm|A_T!xv(SDMw)~-cFhcLkj_(`&<}W5`Onx{
zHYb9X{4Y4uf@dzW4YWOpl-Tul7W<3b9zLV?18Uknif8@PEl*>NJ%ZLIY_S~scxr|D
zlRaHv>31fRV|(y{`is6P&2mgpN0nHWMbzWM2Ep&vM$SQ0kr-#KPpY^hI+ViUnLFo)
zXNELEzkUKdi&|!X!!b&--5XqkipWE>A2m6&QaK_nz)MjK3UMrUBI`BrJ@~IMmtN1C
zVngH;(cPJpvf2JM8Vz@~J<fU#&PAtZeXxI5?@)QT&2lE>pcE#_a^BEb@P`T;R0TRj
zosQ1D-Ox~?@nWJ=rlK~rmk)J_#b;JiVk)Zk%&ZaHEM2z;)z4rH;5VqD*&{7R{$#ys
zZ@M#Z&c51~7pX#RlwL&EP_t|!qgm4Km^)NV-83{6mWJD^6^M9wu3|=?M88D_0;**U
zlI!_yCoqzqj3g=5OS+VGyP1~m)E{_H>Q~8!P~Xyk^o%S}o>`6L^(t$kyGU!WF*!MA
zv=)b32DcN{!8ZDDu{gUJye01@0Ae0E%<(i_0-uPkah9j+1a7m%mgfFh#P`rUGT)vR
z<EZiAI@urFg=~ymvp=T}Qnm0=x?IUaRpd9(-q;oQ|1z^>rz1``oIS0S0%>X{5w#^!
zRx=thY0Tnx=xlITG{yQd`Jy<S(1-RYsa6--o*u3lXukp`QHx6+kqw{%DKAo<aH$uh
ztzvKdAhtY|s%)S(67OwRc7t_7uwr<=O2%5D7c-%f&AU#kHY$~QNqLx&(VB9$QptEI
zaNaYd<d;+e42{KQYnk-eKarPMcav3pFGR^D*@gIt*oQi&c_n<&xY)i%jj(yv!<li`
zw#Eypt7MLlO{Vji>WR=ZhY79&J;2*#GgO}3H2hL-=>JTBC<!m-Cz}d1eWd5M8AL64
zmtZL|XEtVQVeP}$9j8sM*cPh1?Lgps?481yhiK1{AA;W^JxdNF=|oaQr&dyL;RBUi
zsI6m78O|IpjZ)(6Z{%rII5e7G$h1&Dt6#7<v=-Avtzj(&Lip{lksm|swPri2^M4TC
z4O#kUa)umljKdq198VjDe~0yi!FChl9Ugd<ic&iq9ZQCUAKG%WrbZh_*Vwl4II%Xe
zB)BYu((Sy{)CmY@vy%roUd<5-qY?KW$2v95@><Xss-TbH$M)jTzl=)&92n9;%~N&h
zoU2mUjvfpDY5N$w8hT4NhcZbv6s0yOe`O)oRH2DHJz7Ct6m-Un;f+Lf>9i`>hp~sQ
z1Ih?$IFo>2^?$-Qpt;0q`vGeeyUuz_e4r#*`%_)yB6ygVvLusRo&Te+GrfbpSYfD0
zCOtLKcFK#AeOOP&GusJfubdE+&|l&2=riiR3WlzdeGC_nkHqTI2Wn&6M_P_1Q?HQ4
z(s+M2%bl{Ca#6Ve^}$B*U&zfn6C>Hk)mVGSR`RzXpSd{FU5IRHM#cvdEu*NvtPSlc
zmZ4JB$ds&|!edem#aTu$y)4U#6Lx#(DYh}PY}aHef$bPNU_Nebnfcv*P&gGelKI#y
zaZ98h`p8j-{5>a;XoE*ALv3}fm)${e6?uTjj$D?noBo9EFrysS*nR8&<T_Lh`*^t~
zm2Z0#J|mOzqNts0Yx~z5E-b{1WNy(RD@%UL_yg!-IU;Xy;p*9v!*oB&9v)yj;(cv9
z8(w4`rYiDGY5}@GySjf+$ignMEHw`)O~Zc5wbTRQQ{u-+HCuDCl&}R+sVO<Y9t-!0
zo>1Z=Q~5K*EZY~-ApeTQWxfdRkZ-dOEdl;E428-|U9hR}4(S_f=56*hh(Gv)>u&o;
z{iZC8g2}T;ahJ})3a?D*A)AFE!zFzpFYVL$KRjzFOZbn_d!i1N5V>acv!G=NnUgt1
zn~+sW9Fj#Uim#~}kcXg~>QCNz-tnCHoUxh+M3zV9tJTxmJLXY)BeKm##sg8YJYGY&
zQ#2!Nx1`7mrTgJ2_GY2uvGHL~OvbY1hV}}fTH!_bcI>_NlkP|6a9*LvtkdBns=H%Y
z@NDrNEPiKC^e+^V_eigzJ;M#K<y2mzlXZ@GPAw$+%CBr`)X)+mU&7QQb4&Iamc{BQ
zpCf0YNugEhIkpbpD&j$|t1pm5$B@WQs)Fhctit=2u2XiDUW!ffwanZcM2Sb-$=DU!
zY(q(KvVDvMh33ZU)@~3RXxU9H!Pnq7#S`M2T|dbS)-l0akxXW-_zyxwj5aR3(pF$Q
zXFVQi7H&=Uu`~*w<NIbwaz}6p+LF3xt*3Uy{s#6H?L)T;wJbX1Luiy^YP~G)8?}!k
zH;S>Z<hcQ-^gWxP8etm{B0MSf%=*!~Iy1C0*S?-eP#1=VQPEgC%LseGKFF4#;+Esq
zal*NsS8Yc^f02o?vC7KGVU$BEso7cUA|FkEGj(FiLPv`ahZfuS5I@NT%f|3V(np1<
z|EOakWF(kjj(tIAG(EU95+7L+JI4Ob-jo&FsI0r#I=me1kq=SbtQn44jt%&#5<I9=
zAHb>BdtyW8RcJ6#!SlkN$2Z`6Weo}rv*yH_hhMU-9X-NBBk5#n#F1IH!ccEmL#`x$
zIIo*?ZL5{l!I}6_Lz`f!oJGbLjF)O-SMfp0@A6E;B%}j-oY*EG;@q|t!E||~x?kGh
z&%se*fE*?C$U5w#xDs2TYz!YK#^4rT58}G9%J%x;7&%?`Taqo~kvGDk9j8Jh**bJU
zt{FDStN3x&7PjeF5s{Dl4jI(0a<O3mv6`%Gd1LIS#=?RXmgit|EX~M6L6~}tuHnv0
zlY<%IQRx3D&hQtD6y$~$`_5s1spaKyc0gW1)%Vt>{&wo?@FWWUryK!>gt}0hL>H%}
z4#=PJ55QpmN8V>nwI;xjvJThCT3LqLS(2z$$rU3X_>O#qLK=nCs>=D|LDqhBe@TEb
zX*hKR`K)$kUl-eWm5|Af;jflZXeK%YW~h#;!&1jSgdauqQ#{D`P?BxG(XS>_4}tMS
z41cQ(k_O<%#C<UxHN(CY>7{)hUJ%_%zCo&bD}_M-04_$G1@-C$c>}m#dZ%5lwhL88
zK1<c8v(Q%0s7Qk48u%$xT>K$Cf+{7Oi-wK$z$WHPfP^gbHW0$-7*lhpA^r`z1<fEf
zpjX&C>^7{ra)(HgTiB0Mb@Z!<&5@2|kyukWEME(qP_7#zh=I%jJ4eRI+o6X1yx;}a
z=^h4EHm=YQmR0WuOG7$S$R?7Z`RaISMeva&9T`uS#|J33AS3-oaRnEw@8zp_A3SKC
zt&FR+pRlG~A%CeP^`G?zvcS6Bbb#q0l!ix?X%%MSXt2<}C|CvQC2jFfVzQ{)(f-=@
zwi2F48>&5sd014fimn2Ws@+u&>kd^8`oSgAxZp<1N^20PMb{+0sFm4d<xr@lx;NO5
z+JKEj-DGQZxO#!fvL-?ORZD25uqZUta!hTEtfxLI^KGS4eM<<1y*HRv^cUo3=qETe
zY6%hidbK`~K-Yj~*+*-u6628uNFlm02wOtV!L|m%0G=^cbbs*N7N3x#)y|X!*(^ll
z-quz;0$rEf>}0mFb+YnKjbmn}8H2E#8|lD50eVo`q4jd@=tuYrH3#|-K4JglPQ@!*
zClXx}hK469O9{f@B}PZ@SxT{m>V7zcPY6H2Yi78}TH&VZ085SH2>#V}nQu&vP!x1%
z*g$lZmPwDyA)+9#Is7^}3Y)og3b)Q8sZk4vYyqpPlkrw)AoxXIPWi|^-m$@k+jdyi
zQLlqR;iRV>_C}swhK{<*+wzN8mfDYX(LA%<f!~qsy|>ddkxr(4>M+$}X%I9L^8-bQ
z5i(f@*rWXL@Df8Wxs|t+Th6p}ZG*SdHL=O&9%7v7gr|qOG4k0HO5JB9{4ePKMp!oR
z`H2(h7*NF%V-86#(bMDs=L(!<)&#4|%Y^Cnp_cZ}AE6tjKYUe~>E2ufwWe%;%KR(U
z(Y)7JESN>c5$#zJ+9ywRH{q`_SwI!$1@WIBLSE7+x7ysp)Lwc(mk96u2bhEA@4=?J
zg>XFHTDT-n$p_dY5($kaP8;&IZP3L?n9Kz4D#f)N<Yk0M6={opa#5it7uhHs2K@L6
z{}MVCJnt*Tnt1<0hq9N2hmIy7q1;3!&`bFTObM5U>u6s5FR`Qkl;sJL9*PR>jdzrY
zVM;iG+8<rX4i_I9y0e?)z2Xses1!{@6dMErPqGGDYH{|I!@xhdR_KlaS<S#7+E-zZ
zoGH{(Ub;|7<4z#&%bnn4(MeC`&l$=%jBb)RN3LdCWlmuOVmc9^hw78X#+Ii$f*cK=
z#hCD*&;zoGaja0JY~VLJJ4mx|zg*~D1J6OHv-+KTjXl+4+reUc#o6A3Xa^U^%9gA0
zY-Tq&oSFb$<THq+f)Xfi>!LnK=5QYLIex>xN$BQYgI)FCD(Zmr*7sQZH_9hDg`ol^
z#C-LA9DajsW?F~)iAVBS@jdf9wG;>;ozMwhGm0_m_%N~5{g!bQR^md)W|5cs2M6(e
z_|x<ke+hj`>Onq};vz@zs{B%RV^J~t9CzydLT{#G@*l`KTvE(Vos=zu#9!bEys>)?
z{{j6S?GCoHZsisK3NDFAkUtQaa7X2#Tnq0e?dNEwIq|Re3?`x<h#;P=e}FTVwn1IV
zd5w;pgN*`B%#SkdtDkWY(gkh6UzU0Y?$YtX7}FE3zjz5cN?r?hVvX`hM@^<Pd^XR<
z*MOeDe-W3wk6fK0H+DvMl_Y#|#0K98Q#p7T7@=|W5S>p<BD!#u^!p3Hu=#SPWeSnD
zgVrBZLP&X`fY^!d#XQVDz6<e@m}KyBgSZ^U;{<_PNRgp6QXRRUK_QoYw*x*yFCj~=
z!;aXV06SS1V`uuKv!pHszXI`rF|J2MCLV)xbtTYlxfi}3E2X>27nuQ)O`0jb_KNPe
zd;;3aRBT;>c!|5j#=uJbf6Q$bC6FRY+V8m)uxl1!``KyOOkz4al5H$*r$N-G|6Pic
zniS_E1L#q^_xqmfD#}LU>55JT-}U>U24Fj6eEueMummVkwg$QhE0Ssl63C_a0M55(
zx$lc~L3@M0z%B7rFdC5tflfll?fZZS&<9>a%*Ai<am*KXu5P7JMVd^XQ4%-`uch39
zY65NHJ?tlgCvA{eNju3U2z}^HXjEo(SEVJuXCxuxj<kvWA>T3{MN=?}crLDIBN>b`
z57;l?r8^KY(=Hx?ucSWUT4cOntgu@*$S@U;=Wkl_GkV~Ov2mv1!J}9Y<5lkgXMN*Y
zr44i)>q}N+7o)B96@hsfmx2G$0Pzp}&QO6*+foUrF1J<^vB!p=P$J%x?VSyxM){iQ
zvAZ?A6h9!Z^MTUwbdVsG2#*o%g~tJuAD8bft7Sj1PfFup0e9IF&uWl3u`MA1Y4T0u
zH)I)_M!;e_|5g62du#PKw2nTg-D9qeEOGtGS0;Y=t24cc9G>w{b#E6(Aa}%Kc%S*E
zk<@bVZ2lE^*Xz)=(6%-;GkqaS%#Gxg`rVcx-e2%#(;M_9GCj36+R)fvu7>~MYH9W6
zO70%AoOHw7Q|g0m!Bg-~{?)=_#YZN<ot3ayBV!Bw6|ITB+x`yhk9RKmh}4DiG9%6#
zfwqPL%nW*oNt@Y>!-!77gBdfB-&hmhg&tt4j2uNzqB+j9m>yXq<P}aTQ)Lp=sn|RI
znQsR3Jh_0p#2+TkDJfWpJz8MHCi@TbHe`%)Tp42$u&&}Tv0eUUtha@VymGfy*75Tr
z^@M3mVe}8L9<7IJ(8flY7$oQT3YkV`1OE$=0eePwW<B~Rzfrh~bRiy?pThqNQ9hSz
zZp_6elS7R`Q&Hf9W4>n$(O6i<S+FQ~*7MRx`5_%zd>-uOY*_rok>Fp-$K$EMX~TK&
zccAgsQFZ?3@9Z1mGVwd5FHFT<{k@ZwwR~-F2gJl(5ITylTw}TIq24U%Pr-*^INr~A
z*1W)QgU-n6TxW$e-nTtykuQp2vR3RpJT!d<w7KxCuM1TdKB*lJsOvt1%c1AuVN53#
zIZuiy0TTXK)Q6hlHlecEhkeJ?gl;Pf#PwJ*It;04%qn=YgN3K_D@9{LJ>CdEVqev(
z%sG)N%2S~OF&#+@ObO=k&5(I$GCkM1DDQ*6JNg!D$<|ReDfLTs3NGw7sw39OXYdcn
zYz@pu{kj#v0P}ToEnt^e$~}R7f$kxl`8zv3=Ob{JXv{ywmWz*(Qt&N+qa#fH(4pSS
z+Ii9xPrQ{UdV2_DF`S3@#}b@H;z;hh`yck9DM)*;3+N5JzqFseEH{$h5U4y=ndZ>}
z&)IiCe{m+bpSZ@2DX8Ucq_sGkp%{Vie^{y#U$B+l@yNWSjciw@hGB!aO%50uf(ltE
zFURK#L*3KMoP+&x7qWw;sn#rX;o>b`c9YkmUsKQ$x`rRZbBQ|0jP!SF+vo!Y<Hf`1
zI?0C3)OM9pxt^X)pq5%@E=0ORQ|U#@u<WnK`a(rZC*~{hRGLf-;s4X<xoL==-Glgf
z*1Q#}Ecl(*rN8}SjMe3ax<N_-i=^)iZIM#N%ib`sIM|G@f!#%CE3=EP`yTI7f!@9#
z`VZEZSdKrXi$rr^4wol?WbVuH*m>?Q)E4ifWYTHk3+xD(VtT>3Ezy)p(k=WGbWgrq
za*gjG)MsjtQvg;h50AuKf{&q)0pQ;9ZB4txD$;#RO=(`{TYkH)5=Y`ez6t9G9>W#<
zq_cu}(`0q^Anm{*>@83i*-CT+x*{IoEyn}9k!<#CFpgLrn~%H~jumW7_+-xJLcxK^
zy1-21R&yLvHvschw(S%yL@_f#k3c8-`XsOBFUe0?yLA*gpWVX05cJSfCE46n+=EtG
ze;)fS6_aN$pJCK^L%$R0E$1Q+1q0U?T!C1Cje*lpV`d%L9gm~Wg9EUo-f`SGd;qf5
zEgJ@TPnUJAo9ybsJbWbohD&sv6H4XP`W5JvvMw;l0uz0uPwoKU1|1^9^gQvdJV$v-
zag0PT&^681q8f#xz=uG;VAJ3(;+ZNKYO`;}J(jyry!<e*Oz_ZCok`voNH-{s-bVf_
zv_M)1zL&XJK|CQ~MJC~;d3)uy*dy@^@C;pnrh8zdw)b~*ojVWx9cUKniY&l86HCxJ
z=tUHE!&G-<gSiXV0~W-Qe8AwQ+xdzRRT#T`ws9w5cC|5+d?&g98IS+X4GKPV0>;_s
ze(H*V$SktY>(M&tTzR(nZ?O{7(EXaPfb2$%8<R4s@#TcopbET(PS*P&f0TZTIms6T
z>GBM|HTOl_Vw_D3Fs%w`h--Qj>j^9bH;FH?L&zU#^R)+nNa_}TlXjxAT<tH`2!LoO
zX8_&{t<>!f)aAPZ(+LB%1jyif(MN@BV|B|hz5?K}$jSLisi!GFR9v}Zxo<__8MFWz
z#J!?5>^b3gdLlF!Ye_yfX}6_gQ~1q5MQI4|8r{8PyPTF$RdWOB>ZiSD3#~|Q@NZ%=
z6cttrXW3(%D=88AEqH*qh<vsT@NDud^|HhR`IOn|A0quvUyu6Cbu?}vh6<IG`A9uE
z&+;m8REo%Y{uB%$PU#ggAL)a)HTvZuJ&X4+_6a?9m5Mlc6)CT@@m?^bYJ1Ab$sKlQ
zg1CMbQXf9RY+wia#t~ngHPOC|TU#7hs2@&_X8Q7-0;_zd*(CZk+&R4#gtAjTRgJL0
z$!<rhBT?WzCnJlLewjHa?0=DjC`-@+@lW;_|6CZ#EDG+F=Httd#-TWJoVJe1#9T+~
zBKgQYBBC|X5%(D4v)P6n$6LZREOm(-@_|`2Fl_6BUFdssxj&f@z%gbg_a{>qy$Wvy
zE1SO=5ROJl*%!rAk;}vZ^IGdxF&zqGIY?oaXvxTb#>Xr=u1V@&#&md`5%Ul8wUcU7
zHv{Q>v2c?Pu_fi^$Q=q#!X?0<Ks%uVzY=`w9q0@|YxFk)?Yx}1p}d?wh_&@MVa}Vr
z=ypR7k#W|Rfe)SxesOvUa@yI67(hmfk~5-UhE9Qs(R-!~s?R(eErmvSHj<4PmYR)J
zP`b09wRXO(h4!~V%E@`)aDE(n!`wFIL(nU`!2aY3V2f}|3=td(n3DWEk!iLsf~D*|
zHz}xxq{~f-_WVn(Sh_2LmhEszhz7eMKj7|`ZQ68A4d}CWtN6%SOYE6iPpn2iMIM0D
z#J@E@M?=&=ymKSi0C2VWDR~*?{Fe*=<R{?AkPNV|{u?&Q1hFr%;p|)PI*#HeiZu9Z
zgPl8$CIiF4o7~#$gK|}IH+WLlg>O;#ftjPemZs_2!#k8s<T-h<h!St%e5DuXVE%{b
z@Vde---RNBkm%hb{>#@Df1xR+aiRm=2Nst-ukQ3G+KtX6UINc#I~pU-^3{-@=rsBx
z*a*tyULZeeFXR12Qg<I0nRw}h>kR*RL-jNi>4UZiuM;)mH?psBIhKatp}J5D;w=Pd
zufqG$wH#?oH|$7#gY)t%tQF>iH{m8Qx&CeRt#n-IYM3ZaX4)9P8YUx{`vQJce8iiP
zL*^c|SE|o!aSH4$&YGsg_|(4**8mL(fi<ybo{s5xN}jbXTM<r&E~9Ogmj(YOUiBZ~
zKS>PV0M%w>M>orrFfFaq|C9%Kw+qXWGs26y5A!!=R*)}izJNE`<l2=z{d_s}pU4X7
zF7&XtxiA<xOLw!*Hq_!5le^_gd{0cLRpi0=Onvp>{j!~S&{G3HMz`?vCr3ItO>I5r
zzatia)46x}2ebu$9;<3ZBaSec8kbnU=yx;^Izq27e$_ozK11^g-!NIeCHPo*2&7S(
zx!Y-%(uc$j3Le~t+zKo+)l5x<*FizEBE>`P_!(TwDv#)6f%e|!ctbvg^!tbK+F)bu
zuGHUuSxP_%$>TK1ONs}uL($88D%M=91#fvy21xR*Y-KJ8n-i-6=UlU)F>*ZIj$B|I
zrzybM4cVrkubHzqxP!aEAMx)lQ&lu&7usVihm2<yvq<3}UlX<>xX#zD=x#tjcK~{J
z2u8|Te06^quEu{h)}h<+6~somSXtShR?r?h!n|Y_CdKjl44t_gE^}Lc@)-J7TGZ4Y
z-n*%l5HH*)iV8DLySG~l67|D<{~ATtQML{~J+Mmr*LMTGQmm=%(2rr)7<9-g;FKUq
zT5KKBSzhS3d#9!@V(SarSyKN1S?rJFZ^BpjY*^<{M}KSz!JWB3xPj<M`1H!|hI`N(
z^c;{0KJu?(hPjWJhKSARl)AdS9<ZK?XExz`wmnPxnLZUd%(R2fX)F4w(Mynf&>~;1
zu{t`!nJEO*=kh;w59mp74mOPFC2hu<c}5ss8`jYfJ&Lad;rvHq)U(!G*}F>o;jC&r
z#Z=+@2(#Gt5G>MOYv9fn#obY8C>n%+>4Kt#ya2p%-=OX61fUPl-KZLlBGm%honH)3
zeV4p-fD-8ddl?A?Vo-LSRnE`mP4rOkNb#Y-KKcQ8(KpKw<;z2=Za!AXT@rtS4Z(9@
z2jDyupI!lYk5yUwoGD5j2HR*d&|XtggwZcu=lHgCP3}e@%}^P9Z|DUKK^7N}W@j2c
z!)cI?aWk8dG~g1`1J$|f5#5=eKn?M=F`-~Tn{U2X&|Qv0_7)YoPoWlizAM-Jj^D?+
zbjLiWkO_t>rUW|IFh|lecW5iPf$a#igD22?@(8{upoP|I>S^F@GP*_6Q+!|?>mdTy
zfZ1Z6ueY&6@oOZ**_|eVp{dKcolr+t3AUUqzpWyYp7xXe&pg;W&=@aFhIvgsv@p>q
z{O}J!Uvc-Osm!~g7pA+?3g1V1t0};3M;OC0ewe2d_?bSDvXoyLm<N1z7P5Y`9op+3
zrLF6o1n(+zGbKhCoCcM`Eb|GNMPFewft`VJ@zcFB_}KU>d9M6Iqk&iYvc>Y|En*L0
zCV$cmn->cmytzOFBjc^RJsWDoH+G*e9|y-7t7v|APcn}+_|4mOn6H63iyi0l`75Rq
zhPCupua02?2kGI)Gya`rduz9@LK*s>#O>iNYz@A#VHkgkwg`=cvAVY6Kl;P2d&~z=
zhZHkQjGJAFz*I9MWAIDpfzB+wPR&jE=_^kk<&H8f<YCZ7{IEZH;|_5>_z1RYYC{lL
zk7>a+b^1`ucwQU<ED$xt78>Bz%2&275c+x}E=)`|b%GaY9(#5;Pv@tA!;wPW*s?j6
zKxadaaV%EV+1F4V+(TbO&Y|PE`Nmv!s(ZPwz1Yy(0K$wHHTAp+=-=FHqd!p!%yHf#
zR%!lp^`H~o%YgUAz3V*H++*jlO8~2UEAa)pmxyE}@=f@byqa1F&y$urE3xNYBis<Q
z!2OfkYpNl}tmF6^$QkgSU<IDT71lt^EB`>j$Q7V^x*4yUdjhyv3m#B?cE)JKFX<CL
z**AxV%qI4C;xSh*@hN?ndyPc7?euCw<ttc*`i=N}uWV`r^)vStrlB;5FiVi(<^xP>
zu`9VvK<nHM4MAb07g!yKz}e>hK=Y#aem9WhJYhb}pQ@ffOeQucN1QyrKq$j_<Y;sP
zo`c;G_Zo}ocjoc57QbQgql?&@!rr=%L-ol%+Uc$XPFG4_x>m;ioYu4z2-w?l2f?M9
zf&61btJEFre2Z6PrFC%4V1d7#5QN77<q%Ebd-xOI0W-rYauEP@2OU)2gR)>ze%*)G
z_JIlRowOTDFg~Umpi}6U@S1HwZY~f^$|mC+U!_F_+X_B$>$&SlL*Wy06}o6R$#<ul
zZ6D_s*t+mQ{ZMl=eyL?|;4tsVs09uJb-bSF#g&Smj3?PR(;~waWVBz89Tvy%Q~3V;
zBK#ur*88F=Vfb6W65#_y>}?<>9NvBl9BA5|HjSSq3>PLt*<22IQuEpQ(>x!@V2mIo
zv@}6{b#|aI&(t`;C<2%RU-F*hpK@!&BJndoA?<+W^lQ{YEH0?z>tPOf)7jeotm3EW
z3-mNFSq^|Vgaz8l=mEY*W?6>DFqCiOED--lPoOP$vImf!={C5o1jb<b;B-Y0T3a7E
zu0x~gpWH0AA@>Zgf+j(E#uKj6v@@D@^p?PX#vN)lj3tH9Z^)mb-n<2#OSWaKYo7zR
zlCEMQ=}3AO+}>Q5Gk`iNm@e|8{bMuE;ZEczTgcYf)(~WQoiR>YUc7c&bFsJXCX=Rl
zf!sB`w*Ia=X84C5RY3s|C<xc~EQs6y)UtWUiJC|Uxn25A-yG;NGnKx_6#6@f)u4T7
zXWb5`XgJCY3nY^Z_>I_j@R)p0J1F^)J|+LYhUGiQq4ZwJT|6hhCeqAj08Rm{VD{cl
zPVw#sE<)3o*Wf9%k8T&B(Qc;4`IX}P=`Xm-riEZWkja(sS5tXEfSzGL!JG9Di6V1#
zcx&La`H?0YOv3WBD*>YZHnBL2>GI5f00)dW(>q9OK|u2ns%V~VUdGmC^G#U^9SvLb
zKIdP%)5Ul68hAO_ncfSx1RI05J>T5@@i_%g`7hArvaVHxEWj#Y5qDGKBDVs4#3p-J
zt$%|(gFEY904XF$;jU^>XTvWs9y>_SFr4w`5hie;@e{|13i~_O1ni5kNw4t1iM`=5
zSS~Y`e~28UAM?+_?>?c{Eh)xr!(JA=#MW^`{43!93=?+jC}>4DbG4@c?gUXAdI8+h
zpET}bugFn$5Z_$*#O0Ln-d2ITp3VGlHo}bLwIbuH%eH}T^Kstkc#P4(#dVHA73C}D
z7259%#C1^aITMiu{(3ZmjbauW*Re5gO>8wgg?~xc3Un6dL*tb8q}zBz*kPJ3Ofqo7
zQKmheltB9wv3=keDu7K4^NF@wS|MxTaljz#DQ)CW`Y@!g#Y2xmS4xkC-wcTJvU8gM
zt6@ueApnW9_*$?EcBG|*7+YEFjE+icomyS^lr%|-1N(~sb6w#&GF#l7j+i>ijJu_A
z3HTCv3jF2LxW0jhfkUZlwV$fjF_+*1Uy-lPZbqjgmEeXn$y|niW^6aqMxF?b(7I45
zJ4|zv>kaj%)y;e8j<#(6fY2IC#Lgiapfc@n8ns9MkD1`df$!{UdNVf~UYOZK;?>GN
z6|O^nH)aCu#EQN&@hZ9p`8!xitcXqwj6t%XQlvMt3Tg~*WQ>JZ=@n3WsgSxw{}di#
zlfqdBtEmR)@m0caAieQ^L}PKF_zB45qjiNcC!=TTKz<Qh!GDG>%kCNb@iVatNP;*J
zhNVtMNBVktk1vPsz%Pk?0v4O<3R5*kPbGOOvzLzd)dsp3^dnC)GtK8!H<yi0^jdT)
zz(mtEhE`VkPNr=V_Cm|VM&f<r{XKi=BLx+`q%a2<#7B`+{9W#zs~)k=eHfWlJkV_t
zx^guFt<4Si#?%OKEz1P{f%ZYq%hJfOoM>``@AMUYW8H%xJ=#v{2`=TnN+#iN!wlp!
z`z+AL*&jO1HNh70EP`>n(6f98#FgR&zA*zZE5sKrfO{}Y&?U|T!W3sA)0pp7ylF=(
zXQEJ@pSW$lR2i%;qO6i~kLif)Vw!=f@D`|s8qkPu5Mv40jMm6E%|P@L6c9G>ai$C5
zu%QZE4DFTQ($ihpupOLW3}?XT`N$ro74O+w+f&Yy2K^=M;1{DUnb*(*_>K9!vzs>G
ze=K95v=wgQzLR>>S$+E_up3h#6cn|AL8Fu15%^8iY0EQ$ewcX0JIzxhWI9v126R(*
zd4@o7&68xjxEDSg{w)1;H!XYurd6%N*|pP!qcSI$;i`1)jF({B^uy3SXpGR`o9Eq_
z@seH2Yszr*4y+DXTfWUN_Z$;oZW8h|Bfx)0X0kfpaP)w%$RyG8EF+22V6_y_=9$Q0
zGr(jali-g09{>ma3sqsg=xbeqsp96jhNRe7;Ukwpd}5Z<tIhZ57hDdopG~UXHMO@e
zpWka%`9#C$GF9m+{RWv0>@(QG2j&UlZ2Se7U<&)na$Cku=+;2n)F?el?BmszGJN~h
zk6a`%j&Q)a=6Rk3c9V7+yx8*|)oCddr~mOi4$Ly1fyu<#+>E4+=svlZX1~8ZQri;-
zM!@}$b+E?!lyzzQh^suq3o^k>c7dl0-=FD}x)lryD!Ll!AvEzBeTk`y1Oi8RS}1gh
z;uYXAyBNOY-a;q%Hu;@Ma#>wEPY)5cicd|WjWA%;=NdaWonk#<s_s8Qhj$3fFKi0i
zz!q?ZsTp*{+ff8`E5X0akaCa7h6$5tLpx+X%HhLxYg}nYK4T=aI+EhKvT7LILB0h4
ztaUsvNEnUeAx_=Nja7UHTq3asi8`yNt~8yg^TV<eDsNKs7tEE>YRGta13>Wv@&?*l
zRwXzyTX4(!1W6*6Acu83ASFGk;Eh-x{>F_e7J%Jg3Ug1Ng5G1EAXAxTQWaMn&)D=N
z?Q!rtxEZ`f|Hof8bnvzZmLgLHGkA}70j1z*V+mUw<dMN_1bGGCGC%W<1#o@@Qc218
zIDK8fLZ(Y#a9Kx)YCorR$(N8N&dc;fPcPaec(pXcZSQMn5V}G)roHG_-(S!naKADe
zdaTbgtT)~^cV#DmiOe0Yz-vPHYqsk8m`a^akoM%@i&8zXn;BPD;T~d9v=#laU>E(O
za2dEmsEgGH&LQ^`Un>t?Hr*PbwP`$<M68BSGpnUe<}^HFdWf77jv0P%R~jYY3EmQ-
zn0<yPX>3y6)K}sPpa*>eNfvqJFgyyWpg{sfMh!FDe9gal`)}IAcuk=<67)=Psc;YA
zrl$Y4a_m1~Q(u%j9k`Y@Fe${J=8x&Cz!-Onw&-5~lac~RM}WkRX&y*Tu&KgFWVD&g
zZ|DCXdYGOIW3F7GmGN^xg7+A++;m|wG8$?EbT`G=SKJ*WiZm!wVD1{5v4er8fu+nw
z$io=<xz;qprc?*fciW$^(eMn2hBuYv8U6kW#D3FdwsH0E0iSsTmm=f7#q=-qG{XUZ
zg2nnvCO6v9SIR{B2s$}=Zs2WVy6j!s8ro7c7!KD;K-1jMzziq?U-V_8>wTeuYRII*
zv(`IM$eY0R0;U$dK?{ja;uw16)+<C78ecSDXd#O<7U<3%HQ$W96Hft^HTB)!-RbBx
zu%`At-q#<2@3OHSxzc;HNfGH>Q`VM$&<&|e(FJrD=DX($SSzJ+ndh_HSDBHdUIp<&
z4gYlC)8g^qdVdO};RhN1rT54@zu0#NDwUf<?`yUPMrsEo4S{o^^}3gSAGDzG(}uaQ
zg_}<fW9P8d%`NGCXopFHaeS=%pX3TiWnm>fN!!CPmRlCfGdD7ILw_1ZBB+tl^cHjZ
zhCGNg=DE;p>2EVv#v+QqA%zW5m!Tup4ZKh*<h}2mvK`fhnR-HV*_txlwJ^|0+X*$(
z6N)=)I~smyhx?Wji-N;M!7#39I`Ri!Wld)eF^oboaIJ?EkCdS&t@tI7ZgOsx*yo1Z
zX*&Z~lz0B?WoW{KH!y57okcT=s@v-8j<Y`QU$gA|=C46^#!$2ru7NE8`|zugLG*Qh
zLrxBKHCGdVq^)Asu@At}o`#8wpuCJ6E}uUP9)tE;4>AP*!spkoV~+3vzMJu@aTYyB
zxWU~wyNQP;1{tBfZLgc(z;y=d%YF-Nhj}FxS_BbkYyH)P#4XSC?-;?b6G_3&@GImt
z(4@T9!Wv<Q*vXe<n!yazt_Q9nV;mjT&+Il@2h^3nvNOdig=zFjq$PJJeVOqqpDLH>
zW5N`q6>JL(K)q#5bCmc#Z3AoN`k3z#TTD0U0hpv~nRZ8i(SLj6VB)+f<lllkPyN8H
zOaE=#Ja9EM65_O{4GV?Nz$MTLT|lOzX0RF31f-$Q-j4(!pk7FtgGT9x-a+*Jq_@mm
z_kU2iAjk%h;jXh><Ma?};Qo)v=g<oYyRb>H>nCBupuy7A-M<N|_~x!#%sUv>yg_>5
z<N5W*gHT~09}v7JGP<Rnf}11%GrxA9a<wB$c%S}cU^Li7BcLflswL>$$u<Ve#9D8I
zqS3%u{BLB3R~3(Wj_GyCoIpp@QKBC=R(nLh!e7<!z!;Z09(4()O|bKCTSO33<FGzr
z0n{(u#ThMq<bk>(Z?(WgX$)}Gv(P_O$n*e-huHUJJlJUNL#EbF1^S{7<&)+v(r&C3
zJH`ErZC*H%-^l&L#xolNgXluBKevRLMNdq3C3oaL!imIwW2W#Mod=CG9}hIb^S~T#
z8FE@IFK59=O{e7nenNN|n81qW&fAar)})q{eF>}iR&qVyb$UC=kfbl{f*HK!y{%nw
zOb5?9U<y)b_#iwnW$|b9gP6588Llg~U`=2jZi71m_{t3SZ7l0?gUbBl{fJ;V9pLnb
zJsqJNTP7Mi(WRPWfmL(~_!Rn-G|xB%N&yZt1>B?Lb;yRo0pJR-H+-QizZ3Cy*S4n-
zq+j3^yNk0HJl8nAwE~UNc&=FN0k1VyD(fjGdLo#od*fg18Di{%Or!~KUSW9%VVC%(
z1&ZJfSiQWj=4#rm=7rAr^d-+<aE4@q^~ESyY(D4rrF_t26+TWK<L?V}_cTf#<=>;{
z=tUWK^ol25_pmSz-s(MLysO({8cBC#dejy??Z9HjSG3u6m21W~D0m0_2z2A-!1s+S
z3-eu#w3ivRAnL7@ZfCn2Hk<Ex4uf8IE%!i{@c)i92TpmHa5unW>>b8xwq~s5CxU1E
z&waZMQF<0!$9NOT=Kbb{NJXsJv{sxh)N<YOc|5OuXFZr}6a*L4O&tsV!s^jA;2~B2
z1@4$ffW_|K@P7X%#_MX&EHDo6_R-hgaK%?zt`q+-KyL3T&1X$z&iXI*0k_b305U=K
zfnDy}#u(F$9&bFzcJr@?(+fg|=0HDowdtTo1{V1BMJY&k`o7mei$H%~^m|hLf%&?D
z{$<H)#OLPyLZO~3ngJ}<rQz$`<(U4+rJ_4VtIzCS;mrd6<Hr;<MjCJuFqD0~t-mRg
z8=}c!ez5J4W5#wUkR-7aI0M^Bv%{YYz0>?1D0RooEy}unX`ntU0u}w|{LPu>QhCFD
zB*XL2<beM)`U*{+ZQN|{b%T}dlU7k|1)lJA`0aqSPT$p32`tC=W;=1`lRtUhdV^3^
z_bRSBQ%h41n8r1QzCu%6M|`(^O8^Ai3!QhCPg`cl;nmH};M-m-J&V!nXY%U|U36PK
z>yUCQe`ww<J%bEQ`h#m6*y^7bFfnHLeYV`z28Pbgc->3G^I}74v3oan*mY2o=S=4P
zVpT(0ptnm46VAa&9<VVpoG#=0Y12%rd~5yR^ZlHWA8nSH@*acp4KvN#l&j?$2ZHPc
zH;gP!d%F3IbP_yhJmvb`^oWV@&$nBlb|#qV$NcNHZYU2g=7xerTtu^tZ|!p?zwnJ_
zdV~LPQwxU*6TOOm6f;<V#x!QD8Tib5bQLn*RO;gISf-}B{WD$V^d-y%U6kJn#`2}K
zshWR*0+&Q*xt^z9f~Tj-9HD)|`kbwSVtp_6mvNJ66<@_Tm@W0rPd&_^*mhr+vuv1V
zTEP&ci}A7VqxY7tfz#@JXd39c>)hqt<`1xmXcMnba_9~+J=nv^tN2;oR^Vuer`MEW
zh}!gLeyVPg;jBAPe}G#^Kj9iVFZy>1T44-S%X9~HQF^LvsA48>AEG`M&XQl0DS;I-
zt*BzUy%XEgS5NAsybLZ4&oqn*9}4ZU<d74T%Ur~^K;H}+NBObb-~ha#@OSt=`9J;#
z;kL~s{!mBY29-l!MF-nv$qYuQyMn8QiauE$9dd=_a1qrMyPvrSd@pdk6t1Aw*Jk6p
zgJ%^UXh-P5?O4CqjnEXgN$nQyBd1bT)NjGZ)>Oy8_@CZ|K@D#W_vQxx^-Ra@=c$`&
z8z_aaFdyWj!7FGLIi1oAwSmw0zR-QeM#1Jn<&Q`M<XCofp|+;FnjL9wTW52Yxl}EP
z#g+-g*r00dCJmu?t2G>@%-(1L@|aJ^8V_z0;b0qbbfgNI=y+&-fF~+Bmcv#*+L}Bh
z@38(5|EGq?1(|F~f2z=dhZmyXh=#Eu^*ga7IFdXeq#%IGpl-A|HI&|*S!T`#`^mF3
zZ84qDAoH1ZF|bf=EuGT@-~r?wTZ_`L&8zGM=gUCK5`H1~0{@4oq*Pa(a*ePe&xmz)
z<WP@7lc}oM+SGwrm(*e8d!em<Qv|n`M=r^nb|3Gx*|2J{5+-0PvG25X34I}&Q03K#
z>VYy)8hJ-SYCqxu(-v!%agIHYK1lyYyb)5M3R%^OH{pvQ365naNTPWk^%8}(LxL5E
zD#mW)Bk>&ajyx)!i#)M^WqQRj<PU0hXqZ*>-o@JmCF{u09N`aJ-S7i<SMqYKi@eb~
zOPp9dg?tfOOSmlif+@&8r5oDLw;I2!Lbjfst;j>_YbcM31L~LX*dgmSEErgB+h{E@
zeDk%`lvo|XS7KMn8JSHkmsi*>qiMXwyrULJ_9%g3E~FlVWJ&hB#NWZk;3{G=cg$*2
zYM7HlqsYI-Tw8CdJ>iLMRbBaQtktM05g$??yGC!MEM%*)EVadUk88=#v8}i0)du9G
z@a4!2e79pZ=OgE-lY{fz3v@St_R>vw6f^?sAS9rb<VV4?ggX-z+~!U(y?v~$T5Oth
z2Ky$zhL~ux=-KcXep%*G_7)LG2;vyE1UQEdAuiw>)IqVsmJjeIpp)f~vj1>@f)4e2
znX;PdusHZoWBFJ3c({RWI@?v50{>;}r98nFV~>$?RE^A)<XmE)^-pmm(oyNEI@QF`
zKi2V)-y_4KGlPWnfu)8Aw3lV-$*%mx&?(i=46_M#(K0`lVmNF0Pd3t(xo<FG3nk}T
zURpX!TFXW1aUhNC1Q&)g!#AYgEf3|Und3|Ulz&sENnT4i`ZQ5Ol|~&{RMyBE8xiPB
z93aN4udScM)2U@8XMjqPhGx3PIm*k~5T-24*;*W!`Pp%w*oRrUFVcD&3e6K!lzP^b
zP%F&JHYOe@r~wcEKZ?#WK8mF4;<y9|5>Ljby{pF(g4^Qmwm@)q*Tr3vKtg1ajI>nS
zND|x~9^7RYcU@p{ef#}3pECLNbXDDZ?m7SJ!-%dbwo}1Wd)U@ExLLhT5yb}ACOn0U
zRQ06=YNLWD_L0PL$7A}0V^ruY-0SE|#Mo{p4s-PgH>UO~Kecb?*Z{A%a!b^0ZKdyy
z=3$oEPDSZADiu=J5ryVDDb1Ly<g3a<_&`?_9O$e7_hUaxuV6EZ>6pr7s(oqRL~*sW
z-PhYO)VVWokKs#S?x?Bi?ENSY$fi>re^C3vI%0@J0DaYF<m8Y&WOr36?H!mA@P_Y)
zduyJc%G(C22Lp|q(O@nd@AyfsHa!VM2FIue9a&&ViY_IJ+~?T|<LsA|O`%otOKF*M
z(AFjZlqU2dr2<$i^b(GR=7*xlF19HA7@Qq!95gxxI03u36qqJ?PHVqIf2OtS3aw#&
znLMs9!FP-|Tti)0bTgP0-Vus$PLZoIr1S64<jiI%<s2!FR?0g|T<AF$ovKIyvYcfU
zv(2_!O9HKA`lIjE3W21OLC)dKws6yMe{y4J02k%F17D;N1cP=~CI{=6rUr@Bm7y3s
zIs6~B5gekH*dqhOln25;_8KLY;&yOS=m2*#DUu?5?Oe1iPmZ>ABPWDDrPOCD<9$Pw
z!5lTmnZ#^XF9Sr=|3*SnsR1ltjV1SNe>xsYL}&td$?aC6smGx!1!BrKAZnB30m36@
znSNOCJi#~~Fz1MgOc8v-`~+vhvt3n*+5W%OuPmpm4u6(wX-=k{!9`qayUFWJS)yDH
z=AaiMKZH4u<sBRvOggvCV+^ht6dSIlrR={r&e^IG{|LQ9^NK#A#Fl5ij_5mjYnTg%
z9J{FbV5mKXuINe(?<p?hnnrnT4}-fynxl)j$2DY99T|=?gg0F3JZH-AER(j_6MXBv
zbHb~TdRk)pFz})A!7Ac`+>vml>xdi~-oY&??nEWKm*BiD%G+J|rh52qOnEUxG;rp+
zVwi8W--qZ><8YMR2yUQ9S{=+JW^E{)oE{d`Gt4nn<l5OJDc)fu_LH9-i^J`LQ6-(h
zPiKQrO88Z{$hug0pIRC81Ixio`+Hz1EDH63BSESXls-Ayf*rOh#ALFZ*2~@}<B2;=
zWHIMC#|%k{58lKsg-6-X6d%&o@(@sjggq;R735*Xok}Y!w~~+A|7N^=eCVR%7WpXT
zD6Q`}r_z#wE(SeI7C5Szwr6C?m#9gut>GkPBCy*-YQHcO9G6_yF-Cdff38r0Y^xnK
zkb*7<`=CQzh4NO~VyR$X8i*}6hvcNB{D$HUxB^U0p6b{x8AB_n+2S4I3vnMwFvo~T
z!@V4r-Q|dM_|>t-#$##DA7ozWqxu)og8ZUp)^!q9<S6NmV@l|$Z9(XWgfcg6p74<H
z%aRpbwLreo$5kj;LoetnAUEZW^F}C5@)6_A8<2j%>rQLvY|0ekHDhr_FwuNuaeWsb
zbCWq6e2a`xEH+1|GW$_lBOB2gRHxELpo>x`Jl1wS)W;q}!a#iJW$7AXx>VKK#knao
znw}Gm4&5RS+s3Mbv|D+rJ%ea-ZFz}6I+m2G#!tZ!rC<1yCVgpNxWN4lAK>bwcB1XB
zN6PpDRJlRljsB{1w;kM;7s^s!mn=;&2-{Pva)o*|iU)&srL_Z>;vdjmJS^xOc(ysg
zcG<97EN|C{8*IgGzbGkrL*hVXr>m{KhGTOE18-1^ZL_>_<WOfi931)<nBg!$L3!)g
zS;CN*k^~-YW$0;fo<K9EO{f}G#6*-V4xJ0k%&H--M10_gJYSmRsH_n`Qjj3$Ms5X}
zkjL>O&{>HoEfe|=u1?t&(rbaY?aqBnrSMB>x^g6?U0`Z>9MsD_l?EyGlrvI!XA|`y
zvng;C<_A!6o_dJAW}F{bqqJ~lQO$!5nRY@WNJ!PBbxK*=5T#e}4AqEZ!#9<i>?>xG
zqf+=>cp41QgNpU)O6e_@V;kgo!>siA_=U+Cfhn$h$Eeb#fgMCz$jamfGv&SZEN7A{
z(sjeG2dixjVII843y#9%*w9J1#+K!nBsU2E6<W`4O8H4eC0A716W<(lLh*#nwzi02
zQnfTt8G3heKPlT`51q1il?3Bc=ikCFI-=B4GQe}tu@y9v7N;~#`3y91+15f^5zWOv
z<!HE!aoeXuD@i)u$FB<&mqgeKRTtZcY0Q*LDXtgtaeF6MW=TuO8D|6Hfv2{uD^|7c
z6-y86FG{h`aCD$n`PL~b9WE#Bnj<C<HJRhei(vIYS>_;A#1YCh@=3Bgv3c?j&9_R2
zvy><1AG){2VRd}?BDT%o<-6MkIeM5f0xb&mIRSaram6v%UKym(edQsx&&qZ!cNlbB
zVg3xR1IF-GLJ7LXG%^4>vO64QTnE7bvP|GN*$k;{yNlz}{ct6d;w-lRTbkx?1fDsY
zZ@Gx)h0}b&a5p(KxIO5y+r;$XB+yPe0$`v>Srm90xFGJ+=~BwDBWxw`+m>a?!D3z=
zn`e>`Sq~P19cY@PoO04~HiTIX(p|@<@F-6&=dOAOlpfCULKkOsQ7*n3a+OjhLC6Vw
zBrAi3T5fNKeS7$D@Lee2N+?(q=F#bvibQ7tv)8d#3GTDq2$k6Llg5|sSN@a7g$FuL
zyKYdQ<uOc@^b~2No-q6wn(pmwdsufssEh*-b`K3ucH%37=9CPEB&hHV=3Gn;H$$uj
zDs0;*MJ68zJolw+tLh)%%>bJdj;rm}s&!+@DeBpjfU^jkPAUn$U`Fs?9Q(tQ@u$pO
z>`X}o2dbUA<KTjDQV@4cLTXESRIspBN)cOvwyW!vS`s52Lp)z?zqk=L#rB8%)NCXs
zQs*5H=z)&9$#JF;>doYPO0s&@)-pWHwc9y|lpQT3fY%NkS1X6NIj*Sja-;C|fZ29B
zJW;uq`owYFlvkQw(p`H7f0M&PX{OawH)XPYA=^`{gR4-3*>}!$)ISPQ>>|@`&0x=f
zEBqqmc;YQyR>@Ge;MurZN+aT%a#Kr|IeAuD8(Lj_Buu(Qn;YLPx|4s()x85iAG})8
zB3BtEMQN@jqNb&e;4<y8#9G_w{B(6Tk+pTA>oROwTsK&yq*Og7hqbv|ylo<A8F=E~
zU(j;f6Z^7YbxYGgo~^UP&q;y%)*jocyLLE$0}1tWNL-VU-*%8blpIH{1mnEjf^n$N
z)j~2lj?2}EjrN<tG}ps$VnA_Nn8Bg)<nJ7w`R-j;Qp>T|sZGonl3ZcC>^LRN2=C?2
z<PNtVKxIM&yFsw%vF#GyjJxfw5y}Z<%KsMZA}-hu1R{!Wx#EcW^n>sS<~Mn$D27UP
zq$qze9eh)nd0eio6vl=+DMrj^xE>l2_#C#lAD5h_26}rJ8&d9sSfF!m4qwnRmj~6S
zp+0gQqH0N8V6HR1xSjBvAH&Td0%#$LyU&3J_`iX8M}%{+uwLq<9&#QOPvjGh-p-};
zPusWB4)(I#R3byE7PC#Il!5kW=Q3_%$__O%=u$?m>4#SipSC3hCX|klY{_?Bu_4LU
zK^ju>Hk6V4DHuhz3AYa|FCI$PW~K!WIpuIUsyI}sbWUIlJ3su9c6$%Y_mH`z7qPC2
zn|Uvap=dbOc^8y*o>7mNWCz+=r(us_6=pet!7jV(%x|F>j|o(!Z~E`WbPLr_Ew7$|
zLsD#x0)BwGvb0h;ZtEc{#5l1{@ol&qXxTTapL3t5yBce$U$`JS8m3EwsX*v{C=g6y
zOrZf90Y)F19vbiZZa0Oe*anxZvGa68NA+!gJ5cqA|5TuDXsH?<J`s#{9t9p(jNE|u
zN%qXF6uRe_U@ufnVb#!tJ}dO4d$LzxOJ%k1K4io{<S(H)%p6~8>7%@RflJC*;(@KJ
zbqdp0=@o8|H<6|VUO74|OPKadO>Y-vSLmHe!!-X9m?7JB>8>Jry4FFaFk@`LZFixe
z<cepbnqS<uc(MH;!!n9g3DtwP%yhX|;W4-~cowuFG7Hz)IlPuI7Z`P2#oNdb(83$X
ziqcxLrfouPI@o}u0~f(PCQVHVwN7ZRrYK*ie6E(#9<~&|svpp;HYe^?Ur4Q;ane3>
zSo&8z<kvZ?lU`&MHJ+<rJB`;3U2){e<*8xXVjCH#9a5bZWHk1JF*~xQUf3YoL;b1d
zxTb-9$Z2JQGuc)%dtJx`_ZFWIL^{sPKS`O+WF~Vxe6w7$3OIRy`a|D4OC%NwD}s68
zPy2A@1!U|KY^n5=;_;=O!6RUU4?G^_L-4tRhRd7!fYSvhlw)B(`42HDu)=YI-0fUw
zeMbEN6N6h#o5%{RTYGo@f#s=n$fiImqMAJ>I6Y>A{j8&D=tOZ|@l)^{?Bh~HgOzZ?
zDTbz&QkkeDurO3USU@#|Uy-Y-mGA8z!u%0BZND5W8|+kU;JTvGLRJWXusT_pZ=IR{
z6yL`*u*9KrfTCUsJCro$CozRt5TG2Ur1Rlk;T5jxN`*il&ktq1%SUdL+Y<WVdNN7s
zte&<HNa-jfxEcotpm*rRFVcF)6vqc=ePTIz*S@NF7V$o$SFLO^ag2S1442AL<v@bc
zlur*W2i0w-6`iL1{bgUuh_*QJkK?BEu>26OsoseD5-jV?(!5}O@UiqKT|n*2%_i1_
zZjg4hkgDyxt?VL?8PuQ@zM*mAw&yLBTDd*kOXWwXzVj=cN7A7d&VBaf)F$^>XDhP2
z{D(XT`wRqnb3#0q1_){wi9wo+SMJ*{1lu~^2BU#KT#J5PbXEFaK~nfCGFVy8u3)ll
z1(_At-l4srgJh4S1;o&xS#xkN0)c{FIEk&2+cG(zX>!<^9F&Ov*(Q*c92sg0Z&zZV
z<-I@N9%HMd02QDE9M<5n;zvOHn@TUJCyV8=($wT*p@mK~cv|Du<%UDj=RmVyTf1fm
z4YYMkQ`dmLS`YjJ*$^54-Qi*67h*H{4YO%^XPnc8zfe!4^HdzR-(mB`q0P{DuE=1(
zwhB}VHgpYil~LC-TSCj68*;9Sne;3$R{cx4sj-*7dHvE<P72n66q8YWgn1+m4Yd{&
za4k#WT1ca$t{}<TSndPHD_`?<VvMu_>mQh8Hv1}?hEYG+er!X^uA7P8U2zF0WCgBl
z{2N81=z$vYZ|NOCs1j<qm({h7Z$`sJ0RQCI@%xG&N{q*q_*{_4Sb8D-xv(r+Xv&eD
z;IPtH%kBP2Cl;SKU&3mjcubVk4143fC8NquB7)8X3867md!`Pk;(Nw-C)~<ZrkcXD
zHdsWv)nk%HDMuT*`SNV=S8hP7f8NXM!4W3XGl?qZ8vDDcJU3M$h}INln-BHWZ(lRn
zMZ7euM4O7w?BjPw>c4V*LUlk{#+aB&jopItHJI^qjlv_`3T6rQihS-*!3*nkhvOTx
zbi6POGQ{ZC1<LtWxFZU^;zzL_86{_<z-|s+*JhK|&~;#ioJ>8a-#TwBb$;7FI)i8P
zvKGMu_)Tjgc*s7U8;y2RwuCOheNuM(Auh;q=w<60G3f8lcY))mOtKkD%dG-EY@NZ1
z(l(*W?(d+C)IKo9lAw7Pw?Z$Q2HJwVOg6BVwlNaK4*CTH#7F#Sb~=&7RuTH++sU`4
z=?H_3B=%6{%x_JzQVr5_;yKk3{(}2eJ9(X_FvKv!(;RLR^2}9`?^Hu`SN4~aV{R2y
z3C<&DDCLlu_D95h`6)Ohy0ORHd2PD>4L_vJA_fYYa?Y3Kj9?Nc>9aMTXezaYzDIPl
zR*?_+YI6``uGY#)+8GSOwn!LSrORN0K+A2+T-1b+=F0+yJv8kkdvcW|pL2LX70YTi
zwhU{Ts-vZ=z*xUgs5@T;nta;6PC3zUj+OXQ_CJXwX25sYP4i;DjK7)KIMf;(Cke5i
zDJ@Wk|3G}D66ibBJ{m{L30XkPby4lv#?o(n6*!#w!Zs!b(l0FY$xj}adQ@{JB#0MG
zjJ2559LQRSk!Pw&N7};3W$-$BboB^kTBy~Q)?#NLOZ`(lLkfy%p<AYROl3b#7LflG
zu46wJO#l<beSso)*PcX<fU}gE+VhTOy3pSpZ2-p{;r+Ok{s-wRr&y|h_^k`He8Efk
ze_~tmOrW*UnD^j$p=X4hoUS}nUWzBYx4d73Mr2=Rysek_hA&4iSg)(^qbG{H(WmG$
z;8y0Fm#Z1vBV?qdfQ=`=Nzc}KsRN#urdh~sZX%h&Xo)_<FW6og8Mq(~^+`y&fnv@|
z2OGvJUGmx~H3KD_s4P`(N%3-C5K57Ot*VWFC8hhODzn)R=pRaN?m0c))my4Ax761G
znbc#X2V4!m;<w=6)F`EqaVrreO=G%PqOgiNIq@079BHXIMBR;4HV)Dfb>{*Dr7Tb_
z#SPja(b!%4W2TJ10ostE(`(p$U=Yo?tAo)*N7<A2o_zvj;*<5M=c{PY^xu5pEpnQ5
zaZ9m7boZ@gsb0+0++*|`%!gIOV=x=BPP(U`5!|eh!Rb*F{adLOxMhxG;=yr|0VLH$
z{1b_h3Y1ay9K0R!7$>Q|3MM|XEnzk(L&+0#8~3Tr>+KawpDXW-vwSlHE8#)qHJ)e;
zpq-(N6flY-jrc0%#mwg_2YwPa<yhsSI9Ki>CBa6-4#vP-$UPx65vvvc4{L)@<eVXs
zt(`^5k=&2rF06gdRXm!Rq`&LG2B*noiFpOhfL$8N)Giz&?lhM&1wsSUBK}V9ICXKs
zk0PtM5ZI&7Q&mDPEDCRvv4Q)EkES1EHHk*n>)DIo6gHDAwWQEh)bsQS3RQX|eaQ%*
zdLnH@*$nNoHK-61j5<?3XvaMjJHzV6*Zdde0IcP$X4c~+$t!|3f{>2Tvzb@mtZhv0
zXRwo9Mcg<1=aY>4O_9tcp}nw0I1209rzxEY9lJy01l`@{4Sc0mWBFhq7o{wb9AP`n
z<OsH<_S?!~^FLq&^9&3y_T*Q~6$^(+n?OtXj1&&<BpNCUz%NrIHdxzRk~By7ox+Q-
zfwPAC$vW7&lM6{Dwma%(SWRrGHkK=SPeVKU-dd<kf!#emu7Ey_uYe8cCbr>ZSM#uK
zP5FJyc2j@<Ep}dDi7ryGB0r@Mro{zDi6NzW&xl<hHle%CVNR8rfEgf4ysA_Q{7()O
zZDfOTW&0T6yK-Dki$4kC*f)VLg}>}ysMAXO^s3@PaL3TYdkhO{Q<D&kGd_a{$dAf$
z`l(!nN@*}LG>jgH3_*SP1Mwer1;2#aX{sjnRn~D^i8Oj5S1q(%lI?vkCc&?(gw<nq
z7DY%NOh>d+%Q8*16QQ0Og<SA%7wf=cP&V-dl}gS4spQ|{BD>KW>3AjL@Df~yd{ds-
zXkr?_E-N1^($x^&z`fL2X>s&eI7n!{r6t-|>_ym>8z{54vop(<hIf`PV$d?)vQ4Xk
zF4Fa!AH03wE;G8PuUfHKFAg<O)JLhj62<QHC8_nyQ)8bX#rb+Xgv>($UD34H`dpGw
zO_mcT8NcBrNePtA)Lvs(ePX(M|FxU=<;r_(OHl8)CvDU`-w<5JmwSg{b;$$X$|3|8
zJ@J}S&HHL-$+NTYtyO2=M7#=`iR(&3L!SPgZykNdYP7C$bPlb6wZbxDH9X=DFjh4}
zU2HPKh(fpUoqUE(S9&uUu|XmpO(B{(s<L~5$F`W*4<bpDJ4IXJ=D4BZWHpugU@8^M
zfkBwy+z4jlCml)PcKlXF<8F{!P2(99S<4<H2=tozAM=CEHcyuxlKMbbg@!$;ugflK
z6Q0)aGE-Cd4&N(vLe->~SYz7?nqr#QeQdm>R-`NWZ{y}bdpc7X%uElhk-8(xsTuSX
z6)7OlufkPRCz+>LVp${z4`9clzd#aG7Icy7ku{m4#765bbZTTK+dwu^_e4j0Yodmh
z9c`&D<{ooX3MIqN6fd6+%H%DV@~mBp0%8#u$@U^IiLE?-%7v{!u4lWzD%&ZWLS+$y
zs6$$(To1HmwP+jJO1vtE&`N%v(8Q-Ihrqc&GBwmTOlryPCob_r_5YbtQi_(J)m1}p
zz}Cc5F}1WSHJVJqzBs1i?X;Toy|IC037?Dtf$|j1ttC#2Y35f#XJl>3E~-0GHQu8o
zP$$XB{+=L5!I9bOcQOWPM6_`%<iem0IfefS3*;8sTrkR1MAhaC;1g**Hq8GWtE&WI
zwInyY9KRSE&eRP!x%tE*uoAFbr1c3AMX!*5<QGw^!3Egg+?MVLuh_>h$;xFW#yJ>F
z6BQxbHqd%ZDGGIj&G13Mg(SkJ9O3Vwz9D8Mw(~`h@qt0Ahp23RK(~@35+NG4bcDC9
z(OG6#$Zobxt)7q0U>r219wE%W+d(%q+tV|_73xB<QjscK-T@u5k9ltS!2Zj=gRNjA
zuvrXx29U|tow6v8hIa}l3oq1vnCzIC1P47A{>$aa%i&ID3caLEVbM0tyU+z51`h8U
z?K@J5o6LPig{PTM$`)5lY&ZQX_M}>X`8>&39VC%d*rmpQ@VZ<|&{=pj;ckLY-<X`K
zqlG0ln(E52)-ecSd>opu&>%%4V=VGzGcD-6z;C~gn=Ou&4a7f64RkyFh&F)z@GnFq
z&{`cJm7sHJig-<|qi+Rvi(WsE&$o^8mZ$TSO7@;xmP#)RDXJUKGh?*<X}z$E%hfGW
zXW~1QzA%Q5AwL5i8K7+P^yGePdDxAC&zc{(PN0T2Y|aN6-tXb(O04Fq9;5heEe&@(
z+pHkin=dJy1786LUINpxC&+MLSN1dPY?JMomg}B0s#MxQ+_KWbBJ@|VjXFyzb+x2@
z6dNMgrMAH&uX(E$06~1k)CH{z&yx+5LEdgf3c<lvwM^s`d67_6S;yQ1Riz)~c(oU{
z1YG1z%1Oy*d!f|OJs{@G{kXq#&a!UX2yBx^>+WlvM4ur02&3RQEpyqAx#h1Lt{RY)
z*-9H~44g}Tv&|LfNIFwcS`XGoEp?m-wm~bYFQwJgcCj=4FS1P62|Q*%ft6Wrm5g15
zbHpX%MtXGcn{-><Xx|+8rlrgOW_oUCnGY-@F-aXYrKN+GPOOifQ}?;8^lZm6GL>c%
zrscG7{><Bp+;{d;mY@NgU`FmZ=C5lwBSG~Lzf<<f59mm=hUacfW8^ZqMEGp$Td+c(
zE#0x#7Q1<x3dM%enP;R+%nEM<=Si3d%k5~%|Dl^;elIOiW+<eP$5TZ8z}wmzq~lyq
zHXCR7WO=1<MDzcDg=5JT>|SoDG8PmX&0FGjE$}~-pVAA7g^W*`%!F~wYBi_9I?6OU
z$It+cHICsKWEy`>8c^I6k0)r=ZMsPwv@G{=;B{!RfhYQj<-D(o))hyRXDqQY2daZM
z1r&E250+~um-BQX+*Sm|@#D--<wE^z<5X#**pxYBScI)t=V&|GpkN$(%KC)r!Az7t
zh_tj)b})N=oEQ;wVDo$n@%zC-=__lqw54?FO66COC07yu3OR*;3gzURpd|2#e-2v^
zf*wlUl&Z|Cm~DZwU@TRJ)fILZ2f|9yF=af4V{N&5)}EnX;+sIjtoQT)YX0V(TCIl)
z2vrQy4RYunJZmdY_f^#3IPyJt4546SrLAF8pbq<t_)c1}J?Qz+cbjOzv6+GG0ffCn
z#D(;szLX(!0@lUH`Hg5j?*jIs6dgEf&s0|^YaKOUth$@^zzu;8^6fo|R9ofg)?$1Q
zJJ`22caqqTYlaR7FSx?c3|n{Q0~4=2B{H=YmM%>vWa+(WzLFfcNWaAs;C1d3(b!{=
zli>>GHh$WUd0L4_@DASBv0I6Y>Sn%<ZbewjO33jl0grqB<DIZ2Qbk$_Vnw%eDp;ic
z&FrF_NW+pHOfyqeYZNg8%#p)PTWr5h$1QjNhB?7w;pU#Q%uKIB`N_RR99gohuj~gK
zbL;U-mFdVK%`I9^Ue7NK9G7ribA1AqiY$`%y35ju;(S{V@QvTbc7h`l|8l3ZXNg<s
zb>)-9Zf(E6hhp@7_aH8Xv-r~~zpratTj@1^5JIj$OjfUnKh&SzboZEmT~lhFDbvY+
z{HHAor6=r1!xOsa<~BwN+zHaDPlc6|0fzIA`R^-@z#MEfv05j@bPw@G7wH}5Q|6mm
z&D2=x<?ibDu*-p-=<9jqBalRJL+paj2Cs$l-f3c-7<TtH#UONaM=i<Mira4f2R&D0
zHr4PYYt`LOq9=0Bm<84()CNtc*%r=hGOsgE_p-uAQV>65J7mCq)b`!Q?m}w>&Q;og
zkHG43C#{o_KLXR?-$ikzFW%{LdAw|HG&7jH1s+pFERU3-{Gl8aT7(gnGVygi`>iol
zIVoQw6=lYCMP|wMg?>_fGE?6|b}D1@$59(ikC}#B&Y*eO`}w)V5abwMR&7K?7S<%9
zJTX)ym}-e&A8c7raExt4+c8Dl3CEyTOB3rkvaR7y?x?ul^MD@hZN)8%LAmaxx`2@L
z<b2pvZ4?cWY3vh>CoaH~LJw|-KN-EG5vOWGcR_z&Ia8wkAT_k$8T5NcD%J4$bUfbD
z>qOomyAT#FFEnL$U^!R-P)SM5C)h)5<*n)&hCS2j=kIa(!Rx-No}XfkOjUTi^#+!z
zv4Rx5nPH*HQ&5}Fg`Hq!_9?6@Pi24lS}5_-ajQ>Q75~^eN_)q%*qvmt2~}<*PsrNr
z2Y7>z0jtEf{7C5!v5V<S*^Ta|`ucb#=ujBRrSWUwZSxH&K-tm9{zHX*G#=O_Wu-6O
zKMOyHbSN07FZDDxZPIs!gYivTPO*Y{HF~|U6|&ivjZPQdyBiCvRKa~qcwkO6uEnGH
z$vG3)axj)yBdpB*0jCq?vVxvNNH+hY)<I7x{(xJU_(Q4)`}27i!d9cc!iko<1?8a`
zS<5YC=7F)|rX*+PXgwsGOQ-pz<T9eEvP75St3u^wtgvoFoZh<JwS1DG!7e;(x(Fxu
z2lJWktzs?zGO-Z$MEd93Z69?T$Q{)8qFC-IJ46@D*OhI=F5z8P7?Bcd`BU)*IWJ->
z3G-zouVvAD*@gS0Yt&Th0;z}63#+JGO^kr0MPJ<m&}YJI@I`RJ5wt?J<8OnhR=@Z!
zmgLJqe{%uuq;%hU-k2swdA|5Aq0!PO|4%s7o5Q6tPVyvo&N82PnEQgZgUMdBV1;~)
z>W2L`m*bin{^PRn{m59dR2M6cA|HBam_S!791H){PLDf@y@+k_hTvw)SRe66(0#~n
z;a=7UygAX@T!C94PV{QCec~1|X-jYOLHGqtC8I25s21kORC8}l^%}KM*x^lt<@qXx
zTh<oh#NaW(BqV@5ay;7)*+)$9{S+n_9;e$BeepGy&TJV@t`?4Zk7+Z5wFS9oQ|>FA
zFPt;(LjR7w%okH`GMj#<t)zoUZQTX`JED_Bp$DiV-jN6*?C~@~KPvT1H~b%L3#8h_
zb7a2vyT2X(hy-dgaWUK_yX6h5XEHtbt!RFHRZ{ev!*2^$)hT!)+6|TydonGwYI~np
zzzyLh<nG`Lq&<elzBx8R%So<~A0tz}W^{&UG-#<V^%jtwiRWr%r9K=+HVII~JYlMK
zt_8Jp_)s-gYUuuEY7`VaPOdrM(@TN^ycJ)DG+EYYdX}FmqjGo^tU1ww`%lJ|R>W~)
zf2IpIwmdBO!u&?ocuo}DNZsPAfyN*_mCtw5FCqM<VlfMuBmYIUgMaFmSX)S=r3A|w
zqJ}BOI@5D5#K<Q60UE(wH%&uBrVkpcVy@M12r6}xcB&&OOQ@-27c+=@u0=M&njB~%
zPR`14r_;l*q2QtTQ2A`$<y|aHz_{9T+-23W<Qw^-GD~Tq&S!mUVNS8e-yrKwX4B22
z5(V0Yx?-e8xr-1l!sBFm^m%z7cPDs9xJUKNqZ3Ak9_e}$?YuRpH(<HhSR#ZLQK!@g
z<p0Qj+-uF7i8u1RqWWAPE+;fNH%Hkm{_qVXY8(C$n`3>Mx8BNBc`(hkRi%T2LkzE&
z^VB#l8ePwgV+Y&bNW1ujo^u62I#v)1pzgJ8s^f~%TQC##bDeaL+|KecYhP1yYK5;n
zy#pmYZ8?r>LYk3I%uF@GRO*|j^x`VAc3)8ZC}7l4ZU7>}SCW%iM3kpv$Ui_t?rP-;
zc^QXfs<DfsLDFxrJP}KGGT+pu5QF(QqLJ}}XjJ1-OOGtSV#;D2a83Clb}57D>5-er
zdhu(h#@H34R^Y2_5dZS7z<&{QV>^+56KT@4Dp9aB*q<2g{Yj(BYvNzB9#To;OMI8#
z((635G*f6qZPZ9V+3eNag;)l9AY8-XHFu)75)b@;iWXQAY@~LXu2_$o?uxMtqxa@)
zCCf>D<g-*`xGeG8CSJ=L25kslhg@Q-WIaF@V%y1hZl(PNxrZA|&~hETwY=QdU7W|{
z*p4dy6?OJ{f;+HB_(A2C?WXCW)JW$-n@Uf&+*K1`qHQ@=hTBdL4MiaPiC0KX=^Hm9
z{2QDUcR4sKB(5_Zkh&};U_#rNDdZht2l+46J?2Ki2v`e*)Cjr&`I37b_ZIfRu9GVO
ztyM+y(e>DHbdFj(yqX+^hv;@xqUJ(aOPX}$@*7$fp);je@KMWAH((>w@%lsPd65YG
z2c|jS*r8GezVjVdfbBCF#ebrs*oXcEO?#4v<_gBP$9s~mn1NngN($8{K5@xvGv<tK
zIki(sl*)s7wx-^GXlv;>8-df{H{2}UqgvBzyd48o?T3|_=nJ_X+mxxvmM7mS&FESR
z#Ux2t*uBtU(1SWIr=#!uJp%%wS{}-T@3dhw(m<X?+JoD%PoOjNp1P`DO{B4<q+5Hk
z`O?_HFmjW(h0;LjDo?UW$Y4gx=>_vx6H(-4K;Pg&e`U9adJpO`58+JjJFPeUiOvL^
zAj1*tX*iwMNu%9S^6`R7_NGd8aW~crTPx%0TxFermpYWzVlzWM|1Dtv-Ah~zn#p~#
zuSr$CE<^LYbY{IqK|}&o_>Gz(wT24kzEU#D<?w&<B=o1h1J}IJN>o&anbWmSb#&2D
z=_j#)d~16s9Kl;*mr;{&JSPFGBAh~INr%u1OoXC2l1%RfA4sw_WNM4+bvwXC%SbuE
zhe_2q+mrw*OK+s9AQ42Vvqg@ns;p&-g;K$2_M061OmZc8*u0e=t4#*y@Oyzxxhy-q
zDV7>kGc-!8O%Dl^$nJa!7{XOj>cH+QMMrC?%BiRuA7Rqw@$^L3EYDoj$DYt;8@2qE
zl|6JqkR@7C7tmo?Yc?|Or>~FJp>Jb<VplddbClpSk#giGbs{xF*ns|Ji@}Z&eUjTa
z&Q;9->->GyKgeIqIXaHpCVd1`s43R@_FGy9o2gc!W3gdWR$(o+9Jkb3TX<&eo`jH<
ztkKL5d0xPYh*}z?wRBV?AU?{UY=4CKr6^>nC(b<>nd=xM_AXpU^p{OVi^1h0cjo1y
z+U$dZ7g!Ag=DA9)rmO4v`TsF*pso0F|2t4GXolVCrmPdW#`N*Wc#BMz@l3uMz@%YH
zC6ig>nanqVq8e~q@$e$OT!yPn3S5dc)x43Yu7(V~K&)c07>rw(=IUPJB|429FYVU5
z<fG_5{4Zn+@`-wg8R_FiGO_`jL%wi&num7Kl&PZLq^?MP{NIUTbPTqWUaK)kSLOW0
zG{Yb3k60Vim1ztNM&@k90^g}h>K=FjJx2A={>vz8JMi!+9#!jVlEHE*Bt4aT^Ko*(
ze~;a6B&9a`LCO{W1$^SFW^F}}(|?l$;t0M|zJelXEilEdC!I_(d=LI6#)iHk3wf)&
z(DsjdMU4SjWH+|9=p{N@e*3G&CsuzA=jo=xH@UCK6UjTMq5M{3e{#J2Zz_UsrB+s}
z%2&LT!FM8>YGm)^Lzrz|o^4F-&w}J(t_*rebM?RDThe>LTmN#dI*ewqKr7*bG)Xk`
z8|}{gY5c5QiHa7lQ)Bf@ur>fu8BhoM0&&3v%rAA5dK2G>>$aw5C;D1krTHTZg=Fx-
zw9LGK9Eav><l+Nl?P!}^6Uo-MgX?`WsGT4^e?3WhCE{NdM^A|TwR=fe?{aT*!3if~
zQ^Z{CnAFZTh5pJ;@|IzS$(#Axa>Lx%sJi+j>6)ubR=To4In6!B4l|7s{oL+^BJvRP
zHE;sPuv_Ul%q01drz4Z5WRXYmE+RF-5ph4ilz2x*nfI`#i`&~)DY3x2<(}$~QrJ5}
z6)+r~6?kmw>s@b~$viDwNLC_JLrcUC>D`0pg?FACN`~!u%?i00o5D(pG_znDWrh2x
zbS@W8L{=j<w7$6Ceuq9!EdoB9mOn*qk~fL}&}(oVw;ziOt-<@{4N7$IoTHg~8wRlm
z>j!6!J~Qb+4#PxSu1fJ>YSAHK{ibj1VR3$NleAIpre2YQpg9tw9Ouo#TdWy5vq(o3
zq3ztY#5#DtKqExpShJ+wOdN%N(J!DX1s{03h)t=Ta1Hk!j$=j%_0^7<ZNYT0rEj=F
zqZxW%5fRn{qE9TVY$P^{UCC-lK5<?3A;-DLVhf@N=&if2@ypYwS?(-qx?M+IR_77(
z0|qToY#^ID4}wl&9y^Y!kx&s;<v8z1>IspLMftxV<3eNcxBNO*RxfRNofoBiDas=k
ziaSiV;5hdprHSy-_t-;%oWwLt0#C&kzEeKTvRl5wU$b8Y3CvYwm9n1-U`vb(rQ*OS
zs!@;vSG^f<TTw>r1MDfa+xIo9L*6{L5!sB`V9bEEEcek&<StxI55@lCXtYG*m{cpA
z!<MBQc}^k}60e=Y9MK-#tZ@o0pa#7q>$){!TUYiAwh{eHu7E4n;doVRL*y8d9>aS6
zL-)tD#VaetLT{-lyAb`s?h&uybxrR;MBsXU4wVypZOu=dNIp_lcw>#IDscOdOG0((
z5^-Mo?camV)3|Iw;AOnj1$LMH0e;7`OdUz=VBb<PwmnjJ)?Z^f-`G@uG-3DDStFnH
zzXT?jB#bsq0hNe;O2q`qnGkc$zmHCURlNV?-P|-yTn2O0XC$quD=U>&P!DbLabqX*
zGJYX{!FO0WXpKx7$U(6&JB{t?TczxE0Z>@@9bcx-&5MK&$-Sm%csAj@XLw+;xSFb<
z#8~E1H<M;@G2F5I2g(Gcm->QwS#uj)t{%h)98>-R9hFOtPf|T38$LtQ;XO7jRN$G*
z_spM4lS(X+=Em`E%6Wy<=D_8J@>qXO?R}ZkRBFuTVUNK*P{(SFLoKM#$KFpeikCHY
zc|G6~jq(j<3%f?{NAUDxwZElXMlMxdxe2brisV=E1!)O1G9AaJQ5ty@OrvYcBf0)$
z72kB%cj7qQk~a*;sA09b7T8S2U<_GDDJ5RQa~kESnb}Vt!;DM=a|(Z5>4_gvcbNJK
z_tl#*HqNPR_D;3WM?%mPc#bi!U#yY+4hJY@vr9!5c~jgws*SadZ5*{7KS1IdhruD1
zwK~aWR7LqExm&qHJOUq}=9aTG)|?Kk;ZN#-)oC5;DFWpYCp%w$BF_Xl9*OJ7b^*1e
zEYm1^EZIeO8(vEIgyF<#<_y+OH(S%d+krRIck_R+w~#|bi<tO4Rfb2A-{6;Vl-fM9
zmj5J>K`Ou0Hqal*3?}Y)K9keM)yie_|HPZ}A@39=6%NG9Snnlfv8D9GV3D}q@ALjD
zXl$MgDiWWA`%PV>VR;AP4P$;>2d!pZqyFXvv5PcIjHbfi8sa7&shu?SUCq6#O9Y7+
zW@=@w<c;I2n7wc-JSe}0v5}?NQ;kwP%o-pzU=c)DYXho^RD#*)5~xU|`Mc87n?@vC
zW)LfsOKKf@9bXLFds1Mc^n)(Vo+{AF1<zn<DA-E=mWte`ArUkHzqBR&Ju%eNO#UHV
z$onm=fi7VKHQPN?9z}d4HgJtiE@GDZ5$ev50%0(eD)tUn?-_Th$Mgfmbod?~HT>s4
zAnx;S;pT~hwF)B4ZWbFjnvesf7%|d*(KjD$3_9swQ-i_PqMqttsy;o5YHD}_-(w{6
zDe#EhpWHu?CUw9DAl0DXdS4?>&lAz?NXI^7C`m>pWG173z+S>LvxARgF)Hl%%jct)
zBprgy66<Pidkh)Jw3DBBX`z$#6N!@-8Ih=?-0}_$mf)k9qc(z^Ohsz<9^Ru<#lMtY
z@Pc%7V^?6H<D3t)l;Ah|c@gS+_Ww7!p{Z&!Z3^X;y2{z&Q|gxhM3tT>>gku%zu*t4
zzwffxgPTJ?w3398T?Q7hE2HakJG48!dU(zTI*Bcq!5YEr9nn^*K=m@$DiLf$YwyMS
z5EBBQh4BTI@;}0Dx+0{uQcb8=m`*&>pQKU#v1}Hy8KcrqE$65V9ol4nnU+)ks#f4)
z<#NbI-JfWmf+k9<5TTl_^$3(zqgD_z*p}EcPm<v`GCz4&mXG`sKLTb4W?~x{6i{F*
zToKpHFw)TnxnZ787Wi=_hTe`YPAuZip&4K;8IzZd_eknPYWE-n*RfkMjqxP(BfARE
z^-bve${6bdJ%{X6lFXY5JHwlBy25i2+EjDlrcvhSSPSASE6Es~U|!C>;UbkAWQ(M>
zERD1ZL{OI?p`Y&A%#+sMC;@6}IS2#J5Vdl5lPURCyhX5LQfs9WdfYvH-3`oyl{c92
zBxJvCyhh`T<Oln@N+c&Hv?0n#eTj7C2iXHLP!~1w%p|G0Xpq)YJK2ilG5FnAOIYvO
zt@C&WM_(>lg=Dcih*NGKa-N742TL}hgV+vChn>s{K1RU7Q=W)Fi>9!n!C<09{*Q{s
zk9!tSNBpZX&E*5iVX4@Ea0dCny;CTnIMz>ZbWbE3E8h#!eEaG7JcLF$P3yp-s9I7v
z-zBXpuZBKmcEWk`05V=*LELY)lV{}j<cBzOUJ-Z0%+YCh58{>gZ+x1G^M6R%1-%?j
z#CYbw-M;^@7|_=;HKT%D1CX$6!As)1lB+Z~$5TzozsLgRAP87zphJar#?9C}<gu=@
z#s$BPTw}Myx8$$!6U{6)-L@Lr<yP>D%Adwwbh6M7Op1DfAIGY2Z>(QX0c<lg1DTjT
z&~8H}@rf-EpA$8$pW*4O0m7Pf*|0+4Ckk?2?*)Va9JQFZjcp`3?gsYRQ|RrXbA&F}
z{KLG>u!F58J&i|T2~zC72#+9hkZOE+;|ctSry|pyZ>TAlV~F!UhqsJRHnkvX@^jf#
zFG@`09)WM><%JAgzu-H*nUvX&)GF+LUXW^saHd<Pv6*ji9a7fo#pYm1qKNDeb`ev|
z2K6-lffz%yviyM*<pb_C8S+WQY)UV1f>+z6`_ezSd!o0<j{Sw8@=dWTyWiVLen#c+
zb-WXZ(b_&W#B<-8hdr{)G=7(D>20ybfjENBsV<(dB%#wKy*OA%;7*bq{QngmBg-jy
za*?GsH&4FFZ&Cj(`~mkX3#4x3!0ZhF`l7eg1z0ho9v=z*PFmv`4;@m9<q`T&I!k>M
z`oL`1gZh&UA)Tldn2F|bLsFFKtx|^m;Ym|$RGGp^bBt7MzE4y}WBD;T-;n{bs+5Oo
z)!*(tf+SdY)W4pXsyiTew4T)8gQL8EK(lHCeLZsw<qGVkeHw)_jo3-6kvFg^Wc8#M
zpo}q{p6S0y9_Am~PJnhA>-t}DFBxWADSvynY@zTLP?il+n)aqo#p-dkLXB1_)s$kD
zOo3pQtQqC~PG$-|e4@8xb8X}_u|~RM^dmj6Y8=7Yx%q}3NDQJEW+1C|=UK1zram;+
zwYH}h<V-+6;IHEcatOH`fu>`Y!TxT<2rd;%hGh$f6Cz~{3=+>;BCr-@b3b9NUNlFz
z2MggxJTh^u??iS6X7e^l&fu?jccPA*iGp6uG#x{t-JgUb@hBCh;<$Oa9V`)s(U>{$
zcfvk#VeTpMi@7><V^4)$SYyjlZY$o3R@u?8hFD8%SBQ!c*c7BN=LW}7bnJqy8^uYM
z^4SzQjw1c7%pG$Nv1_eA*oD@wPOiu=zC{{>1qe#*vUD*gXw>;cSRFCB_xLZsM#3)r
z7;=kH)xH*1r=NIa;VN5?SSc)Kx8ewP)$-N17>-T&s_0B7`99W-J^;f$8_|5nZ+1ca
z6}1+(N2}b*Y47I4oF(K7zNwoN?ZPr-iN=Zinm0!H>6z_s;l0RpLuz|kpnXsm8?*0I
z!ZE(BQiMIFOZa<848PlSf$fLcjA6qCuXaNNUk@dTg>WB+^UsMNTq*e;t3=NiqP1jl
zy}V@`BcaA`Wv629xSiJOTpM;wu`0|HCgv27MV>eQy7~>sL_y>X+0XOan>QC`5H~op
zKZV>#y~;VP<&!qx|8en}_x+vzBiEIGPyE96=1w-Z@EWC9^Gnu>?^3eyYgmum2z&}M
zJLwFv-!sTl9Z6>^#TOSet=23y0Lr>2`Cl8h<4AN<cUL3yBY2n{$o-F~wQDfHfEeXj
zizfJ<!V2`Z8a~hsrct*Gr(0j+Nob{liQ8j{4_cS7!Z?6D>)iy8cn|09_nhXQsLf$d
z-dpu=<O@049)rv`t<Wt(r?Y#7-p1JoT6Es?MZcBvp$pKZrg;8D$!Jo?ogvl~uCtaF
zr4hQYBw)IbzCY4X%9mQ`pBPVJ%S^?({hq4qEZHKRD!9j;_EzIgN$o4o_1HZZ@wU<3
zk;_;YtU122@SLHW8_{Zy3HsiaTjU;#PWfsc4^zdNwqYhO;UFs`n6EZHnyl{|h4Dx^
zb}w2@I)fb%(g+WJ*S9K>5#!>=!!iDU^-DMBo2QUC(%2tIq)KU=5gpGrq7p61(rWD%
zkqUryC&6?WBC@$_bW?GwSOKo^PV@f7+r&}1Cu+`c_wn{l{3P|qnj|K%pI94T9p6D-
z^d?gc*e>Kxbcw&VT*j1yaAGOU-NVH%!Rtub@)PK0BaqQlalsmLpZG4f8d*=Hdga98
z)KYRe@so=bOYug|_lOGr^{<A%xE5H5n4c&biqJx9v#mk&M)oYU3Hyprh?eH|aGCX<
zsfXnzY|r`p2WcdhXS0*`qMx=-_Vn^>6r+@TnS&7?a-IiUp1^N~GyDgA3)~mbR&W&t
z%m@8d@mJheUweL<iL{uCUML14!E#g_EZ-=oD(=PI>^uB-ZWVnytRt2|R+c++TkR*{
zs9iX;n-|Kn4PDutLIir*cR}edzswn!b=F+P+@YW=uP4mJMxiBA5q8e6F8?Z*h_4fV
zp_lN{x@5}<q(xG4tX@BsxMTW?mNzW1CR;mLroyzO8E6@3@GLF(MgEVdPgaFhQk=@R
zf(+ujaMOBFs)kN9h=hj^izBUt27pfUWN_utR^0Mx2a${98tZTTK}=0N%eM^wH}R{d
zS>6!Ock3wWp*KxRLwSN{0d~trno@}2T)>cPn#ZoP;7A5rU1=ee6U3qemceoh{<g;6
z`ouBjU%I}!YxoDBo&3{yo)4HN#QtTh2N*s=bLn4X^SROHbbfvg1rPCi3kK=Si7ViK
z)=8el`UoRHuYk>S)B1)S9;0AgjGXIcVQbG9;_1fj_+fvZK)|IYJ^2FfS9DYSpXYE!
zG}RiZ>U;reG8wsZd<94Y{+;)=P{nkHDwDU&vX`mjd6O4e7?F7q_QN*24|!I|ZSc&Z
zE5akoql6!UWBLp6L&)QBwsg%>PpohLf_}mNCF>eQWEuO1$%RWC5BsxA@?~NwU(S=_
zizTLVuPlF<-};WY`w=M-LGL}T7hxe65r6BCBrnGY(HqV8&<x@nchrBA|C6jKb&=W$
zm3-rk4x|G21pL$wfU(xCNN3~^Q-Q7>+~TFNCjR<}g1m4IG}Mzjm{LKp-_FnQ&n(11
zL$RW7xNKxs7E(wZq|C-_eFbf1#1`8$m84C258Wa56Sukk1nRxPO!kNgvm|#uKR@RM
zJPd1@Q{pqZQ`TTkBcW@qXj+b)g45$4!+~fl@y@zXoa^0SF2GG#cXWGVU!P8JBSVmG
za5Glja}v$>t@P6Q{X!Jgueuj@#7bdD`Wts<Yj2Oma2rLr5=``;GOhEMO&mbqMJV24
z-$*R;jWn9@Ov^LA{f;N5B6lP?jcwx_t=<1r<gagDX)<Xl+;?jb8JF@?KTf`pog_Gw
zf%GfQeR2`^gX&m)%OzOPvM8oE@m%g)r=#hjp{2E*$u11Bo%Hvn8zydq^SHZ7jeLK5
zC&Cp)9pMl3i2t2!H1<C?@B1ii_RPl2<Q$>~sz=J3dlB6=Eps3ki)<#y!f}QYOB3;x
z1kr}nemvb*iq6C4iDBa@X|0|`4~qxL4&tYroBR{XN!5liu?Nl7=?;di)^c!k)<M23
zC3!RQFNjx)Iwl;GPI(RFCewZ_*)uCXI+j3M!awxW*j(Q;eN)sURFdn$#;`@8ujTzl
zBiki^xBdnC-G7Hn@W&N>Equ%k<yONVrn(8Km`2AD`WE#wH^=KqJ?S%WsAnYEmWV~q
z6mF3(=45hV^F7vY|B6=gEg+`G4)Xqn(>?nVBDiF{P0~uGig6aV*!q;$<^Q2hp`w&o
z=Krm4q?@xlO|nXr3SHy`LwR{#VIQ}N8;ErydMSB1ueqGW=VZr%kK$BR<j!Jt{mCQ=
zn`7CmbhNyKdP{#eBq0Gt7cODRqzBRh@;-d)8OZmkQ`X#oG#MVFqd^|s*+5y3V13zv
z{%Ocfv0Zo>6)<#1hFKlhb)m1Vy#KYnFFOn`Z+^-eJQ#N0Xi&z}TVjsuW_oKR{y@4g
zdyp!DLep9OI^BkFRd<qj7mG}2Xey%^^mBwXaVU6{GuC_7jB76Sx*B1s3;Y+hL^I7s
zaf|+{zZNQKRYQ8=YW-B<UpUvRXRD%T6Bg^H`kGrmSFOu$Lgq`Skd_3R*jl`i%ked%
z+hZ1SVEiuqMcrF?(svJeEx(V8l3(al+iR0qqbqO5)%?)bF03wVynnYg$*RgXHPzxC
zxi_)P$yiL!Yn({pC;Y#L&crXOa*gBH4G~8ZWENoN%vs;_o|yqfMMO;tQ^`vF$W4Xt
zCaIW9S2V>^N+q@wL1KsjW;y#dVB$6LqiJQ37Lu68)r(1HE@WC-H^b$6&)@KV-sfGO
z-}C+U#n-r5_L1D<ijv;8k5RV*x0}^h!!49cx#>n^EAu<nOHHc3MP-x8@Y%Sh2%oT8
zUaXbyv*p1J1;H%oExbcaW*a=O%AdQo80V15=rI@SePF&m4h~n7t<ChZaEi1KKL<A9
zBCy(ff(Lb(l(>Fmv`XNmMxexoy@)%oUi~+0x>QSs$+wX&L3-1U0Ki%UjYzhAtk_eN
ze`pRk!+5m}unn|nA$=hxY0oS3m@<dmni$^Zz{IsNPd6{4t8DlE&k+ByxBD%iM@bm_
zY|XFS5qXL3<w7u;E(WiQw^X``5^ajbhl^h{b|IsvdcB=27iQR}yO)AdEY`qDzndSU
z-Hbro2_WIG9Z%?Zmro#3)!!BP8@2~MuD&5I3yz?>+!u&((syK=wMlOUU#Mv!Ma~cM
z?jz)KqfU(w-xV)X1(r*3Ve+uSOV1|1oAemFjr@*wBP;P@QJ|OMh2V-*q1iPz;NcV2
zi|$8TX-g<o>y^F)7IlPhU)v4mY8&tm@OjD`%pFL`x{)R-_MB1QBroHq*gL8iX7LP=
zKcPBs!gxDOl)4-XToXNJdJi)PtCr?FvV3<4r`qCLZ;)zp1n-niIzv3myb7|^+UB4A
zpSw<fXb}#|PLM7x6*5ujqs`%&uGfL!Y@u#(pVf_KXqPB=5gj0j7|>(MBMC$FeCZ++
zkY0AA$A7GsNn1e_lNl<Z+N7`8e%EfeQa-9)NAfY3_(1*?nd+hp7F|vAT5XrtPv?3v
z4RX`7{Qbn&@El|uKMn5E^WZS-2KWbEOScnk&dkPRPn{xn5FOz!2($QiayMJ3K6dpv
z*I@l0B{3mBI(!@{H)Dw%>QJn(#wbtVdr)GlLf6ThU?(Q1sp2`=tBs4jW!uiaKpp1#
zoJDdbUG3lKIupLCOu(wFy}Cy`#F^ka?k0mYoDxn2#t8R5v@mO#>A`M1BfJ4lQBE)~
zd4AD$5d*{|%r3pdT{pPs^Kr)8nAA@!;}26mNH>MS#L4<{DyHr>)k$p=-=w~2EQBwK
zMKXXsE(K0klhwO;ik27eB`?z(Z1d_KigU&J;q1sX%M*u(PNh@PWc;Z8vbrWDYdE|J
z76TVL%G2eXP3>#BYgvh`7N#WB(G9Kz+I#wYuEAo2g$|)uCE7yG0>8$aFxeRBEVbvs
zXyym*nlgb}pznmyty{<@el4FZne==|n*MX_SMJ}yd9jT=nQ#o*uav`|&<JjpG`x9z
z&=Rwme}&$ua?~c>0Xr2;T1Ff}({+Os8j9lk!hZ#m8dKpV%S2)ijAHJBE|<~6u@y2m
z_*8Nqa$XruSowQkhP_ic6MTr2%QTm&Wy5aGB&TTqHCq}BHMjcE_<=d0{YaUC(wqnm
zcoU_u&S_)Z*d3x8Mxdo^wSJKv(UQc@m{`JZM$2Ie%p=x0Jxwj_xNrqDdHT#Vxh1}$
z`X5JJ5VF0Av7~Eu^IvqyL9gHybK%>;eR8&ADVnJz_{S%7gR9<E%xL8XDDq6jzixg?
zdCqW4Jd#n@X*QiH^53*wbB@=y!(j7V;57F$Gh8(-g3_Y>V&nz$ls0m)c!Md|R@nD8
zjCYI)uHd$7r-+m82y%JzQe|XJz_7AzZ!!~?$$s)r!xVJ5ISzY68DRUFB;o_~<N8%i
zX$FIF2ssosBmaat@h-;f&!80U8{u7UIdMtvSEoo@)rD%JTn6?lcU_%IntL9+jHl?P
z%BfX<AW+(g?bb51v!1~Q<Nb5(C-oeeLz=i0yxL&)j)qxkfljis_&8*87*|ntO#C5m
zH26X*?I;#c%GSUG>KU{SJ*T#S8_r>jS^6J)2`>*_Az7+`|B<{3h6_#TIs1C9iz|?B
zD}&VYjv>K6tewsdG+8{!enom!UOis7z*osf#2rDO7&8>L;$S{PKNg?nOyt|>&zNoT
zT%b>QQ9Y|xvKx?<uB`Y%f3D}_;OY8y!l#x|efHBT!e1iRD%Zuq<f6)l)VQVv`07v%
zJ1PEsUysW9XGy)tE8-6E6Y@(m62^jh`(|MVm+J_+&I(p*Cby`84z6TB3H_j6A=~li
z#cVRK;Wo3+F&V7byN=YU=paY+@GBKgDwEUXg{Yk;{3DT(js~VgxM%QLeIS>s6f+o$
zR^{Fq{taFXrE`VGKCCbjCl(@PkCG0_<sV+qs+g}GLD<U#V&0RBWUBE#k$|oCJL@e0
z!;++joY8PmTu0c$9bnUI!(~xqh3bcA?f<ry5?WJ)nD2O>ucFJOQ)G*h8J;LlmP(bm
zaaDSdyTQ20wz01?zNY2ZUv-6<MS)nb(c93NBJu$<%m^OlxBAxGr>Pxrh^9CnVKZSF
z+yx@xJzE@3fp5X5M6SQMZUa7BiII!(@2S%GHK20OF}ch<oSUO0CPc$LdI?`@9tgsw
zNM^uiL0ZIac9eXF)2IgwqhAZO(k1erM)G8y_`3KnGB-{{OP#m){oIDwZ|SG$Ib*Z-
zLX@V4N!P`T_zc1X&jvPWUhF_<DPDrV2nsxX>?Wp&J<eN^=lH8I7}!LH)lM408iMzb
zXA)laoQVEY?AGI%n{dwFfPZo50d4d7<#x&h4;sFcZf=bBrZClMg*~N_hQa0}jP(vA
z^dS{|J`tmjNBx3}o<{AFa;yP9lJOZg`*Wzt{{nnO*4Zl9c5IsMm~C^y-AW!EO7(<J
zV}k@U^B3tvuA#~3j^H8qj*gO3*i}KN`LXyMGMuRwht%u9C+0~fh&A@A5I{DXbzTT1
z2UqNJGn4H(T2A9OvNH5@V=J{Ilw*&CTcm|bt7-~f=22&oWENi~kNSgw<@i4T{-$Bv
z=+Ga;Mx(<zPiDB;+--IVbw*wkdyoN!Q|B#bt#b%@fpb&8x~h>2DyJq8QOyOcH~ukU
zRQASbsurqRRz)%P(BJ92!sgTul~Rxv$ETzz<b>CZyW;m^im-tHAC*O(=dUpLx%q6L
z5Ebb4r!z`ii|rlq8rbPuMYh`a)(<k~BZs23)hB!rv~5rv7=?}`8zjkCvk9SL>{f9q
mxX%w5p126Jz|S;w8mIPbiU&!7Zr^Ry&*U>K&f3Mm0rCH|3lOvb

literal 0
HcmV?d00001

diff --git a/wasted/disco_test/2.npy b/wasted/disco_test/2.npy
new file mode 100644
index 0000000000000000000000000000000000000000..0eb85522cbc876fcdd4ca675cd99454738cef653
GIT binary patch
literal 35968
zcmbSyWqTA^*EQ}EAPJH7E;&_os=E^;xVyW1aCe8nT>>OT+p0@WS$8KPxMYxF@PWbI
zW#Aqd+<E%_3vYd^>JNRcs&(zX*4k_D=+?JukAWpi%q_8q96e@M>P)hcg={i5nS>T{
z?6jG)XO5gYeA>*>WB%Xsb|WXx8WVXwYy8ORV<Pt;30hL1CAnUG7X>X#EdL)@rIzX;
z#4<1WXnYT<4;C3ysO_+{%7`_^zwCMXWjZ&u8`X2mE`FLi3Gc^Twp0vJ{;R=b;^&wu
zj^$EkoJFfRzpbovgZjj#B=;m?&L!q_uu!=yIjMGPWBxB=PkVpm4|4@#JbziJsa+LQ
z#7fu}?-{&tt}b3+A5higilBX{oc>I0PFuy})<<YPSZS+jm?>`Lk5Mg^Qdk@FJeg75
z%2wMPzP-W;`D`AjEc&pwfdbD2ltnzpM~i9T3yy^@TP~B|f}P2G_&feM@mLw0uvuwZ
z_l(s|f6$u~we?qUg>aEK`|a94+UcCk-~+L`Z4ouV<8mLYGoR=pB<63>=76?Yr==3u
zHaLr&MEnfIYi_ElBO3jsJyn)lw&B-Apx*KysDDGGz>1`N;V|qaZF3o56y&Hi!nVLJ
z=3@3FqK-0*JfU9m2@It6YIg}2e5%bQ9tL)T0ZM6jleh{l;<Myo!Tz#`BK3B3iqKHW
z)k9zn?1v2o8<}oouH4Z63~%pyC$6xBgp-Pcz9z27{b5;Qc>*i@`-6~SGac0D`TwnV
zk{*Hj@s+jh{E@t>j{A<^#kIjTWD`>etCi3ibAYc*XVllV*|t=xqa|hiiGQc$3>KyP
zOHwtZ+uLv8V`96S&c*as>*jRu-Hx2EuH&q%k~(vVd?h;YAJhm<Q+qi^Z(66Y^4FxZ
zY0bHV(jvKzHY%?i9fcptS%YPGe)v}G^@B#*J>sbN2S1NpBwo`-YCo%&a1OF9EVWSk
zDRa`mq>-v8Is^TZi{p35l@+MxdRj>v9N6Y;r3g*ZU$Qs#iP{E7HmKl@;cGK}$x`T*
zyvm&GZp6(aZsqoiVAU&cD`J=1$pmpQvxeP(Rn$8aK6k9Q9pCaI`<P~T-J#yX-SQx|
z9=e))j@m}G5c~K@?H}qOxwFax;sf7U!}xvbec3M~p@Doss$ey#xnMGN#OT*%APc^e
zE3jK@<;gJ)Dey%40tgVp9m~8Q%n(k2zHC=G9`(0f&sl(cQYa6n8=|eT6YxWHwm1*`
zB>&F)0V@$bv;?Z8Fu>}^C(G|tuTs)5L1r!AJ&e@d+7?X{cQHTZf!tN<AF+mZJIgM0
zM-?eoxR0YP--((-@5cwQ9l<2)5#g1v1)tBp<oXCB(K&q*KOd%o-CSkvKY9Ot4M2K{
zP*KZJ{xq~#cyTCmnu20Y$)pUmPavc~6LEM#T6VNW1cNkcPomTcRn@~_jODu4Ci}5y
zq;I44LQ8vDM?-<+#weqh_R>!)N8QreXYJsA_f<4QAw%!RUj?0G|3?3^`>Aw8H?=$;
z#n(iWP_ipa7(p0_si2Z9(5);@@Eo-nSc^7Vdq~SnDS=h~f!Hf*5%v!eO{BxqY=-3o
zeI6^P{EDi~pEu?5ZNwgYB82F0Q3Yb-Ucs2Wy|*(cbYV9I%F2zLXV|t{4qlkkTsa<}
zpVfjLuNDcTU{AG>xhQ-#lnfn%mr1Lr@oAQg`PcbFWDvbKZ^FJfI)VzIy``zJ)761~
zn_pqG%KS-Q<(H#*z<_6wuUvVX*bQ?u&3Dc2XRMZK8KbBTjty8kUC&vHyBg)j8W<nB
z@7W*gC8Tp=g88#OKYp6~v-J;!#YpNA`l3do^>RbUM9e|8%F5SU@@pll&XQv-$@E~W
zQ_0k>&}G<TVmy2x&(I6;7ufI|zji6yBK@0X1o}weD)KCay%{qX?>hwcv6D#2Ujkp5
z*^<UpmmCkH^+a?YouW@uAIMnxv@s}ep{HmQ3kw`WK{o9X*MhbR#}@M?iAhwH`#L3B
zuPO`jdW$QB_Wsgre|V5i&s*SjipkE$;r-!$fpHWAaIoB5m-51AIj4j7qyvmd#ruzD
zy~>J$r_>7|8MM(qQ+YOq7V$~yW^@v4%$vpko0NilOs1=Xbio*p0d&zFuWkq=u$xpR
z=9WH4;^a2=pZF<F$UBYS)bBCLAkUoz_W`dg`InJ?Vg^4>s(@w`*z~)A_1U#daR<H8
zeot+q%r>N9Z<Kxbs$fg?IQV6q2HwNcSWW*(X)az)zl%Nb|3@tiKKFOvExT8lUr-wJ
zSe-+QblBQNTPQsC^@J}HyK@$`hTwpy^a8e9C<J%rS47p+8gMEnfSz%u>u#-gNUIxs
zo)Asepq3bo_*1M<xx_#AZKTTS+id&MD7kttF?39`**yM6z@>6LA=s(w5=HwFZrD!G
z;|V5{>uc|4d855UEwxm3CBBU=NbVJU%l9@{(E{iv`3nEZ{(vjB_Q7^qB43Wml$)!y
zsF{457z#@f-V2Gw%f3_0W+g*=m0yEu&&8S#z{a?bDk4kMp4@fvHfvDp#l=J)+IKRe
z1lkQXY9-acZm}=*u4a&&;CU<LlD(+@rmgV0mZKi0&R{{1qItLhe1riQWAN5+5IurD
zh?kM?ar>BIv_Z>djwVB`o4pTsOGrpDC8Gz#cfo{pj(;t`6J7@D1`-oFAuY$m#tDO|
zHOwp<qBf|z*lcTKYiC~vJrQgVzk%&2K`yIJW`K7aPU+{ky-I0$P;iS{i|($x1vnl<
zB~f?O*2-(-g*6jSXq5I!#dt3{j7=AlgmaqD^p5WwNXL$u4>42f*9O(-_oS7a2Un=e
znIfu#+*jXesR7>S942alwT=(g$?kPB^2f*nKuvI8Zt46)WbwZ|$K-2dwr4GK5cMRQ
zFfHldwQ=}l^MCYZ<bYB9N3jQXPgx_`?1#}j>Hzl<*2C^Z-u|HU6;3(#8J#GXv;N`?
zyC!M-j6z-`e;fJ>xEX$Lc*~aejiDxBm)L2j8&}(qAsk}|2u-bj1q%~j`uq8ZDj9Gx
znhs5HHunM^A&TU7ViDPj-$j=LtTMqi6Xc+I+CxKq^Hu#ab(dZU%4R(x56Kg)JN?L!
zt$r3)Ip)hZX@7tgyE8NLx@s8_WE}_TzS2^ZRMj;ds}?9GIJH!GJUpOJg%h<Iu&jS&
zuqL)#f5Yq!-eV4tk3l(aES@j$+EII&{J*@Fb;<@Sh>)JBZNz)~R&JX}?5R~HW2t3W
zV5e~`E8vrb?IhvbCi>WC$}dz^`L2u<pWrR|-u@Rm%LXsgKk&0v_u?x-do;jR8{U!M
z7&n7d*WjJcxJ>F?P9teq`bBV5tuGhycf_;yQFtqCa_wHqIx<!LU;G1MmA(kPmtqYE
zp@P3eoB3jUM>JTBrjam!-wtk2EXg`pQ(f<MauU^m>g@YnY_1;@u1VkEV=4iZ(G~@F
z$wQ<?`q~_R?N+K3GfmC5_mC^HEAmUiZnY1^zce(r)s~llaptv5GH)<m18#dgYkN>l
zorf)!qjMeX0Fj3S6o@hQJT%yPOI{&|9evVoXa}e`_8Z%jEU9<(w;^Uzjn>u_-i2^w
zq)<aUD|B^i_s$Gk6ij#}?LpVYurH1B)2Fh+Ok?G%s)|dgP0Cz)bbug+@%NRhvW-#n
zL3G{tRHkjI)gor8BTs{OxTBgFs;W)KZh^L1btka)l2)LG{u*R~?AB_@wS(ok{&bpt
zhxloGSm6wNSFL0-qxn7&B4#Fen7@zS(OW2$t1X&jTOc1T&?bSk;BQpPl>yJ7Q`~n`
z592VVIkwN&2_NQ9Mw~5RtLSp5MpT~8qW4LE(6zyI@hMtlE*mh)Eg@%_AoJQO949WB
zW5l{UHkpRgCDoC>CzQZXkpFQn<|d%gdE>MWHqx>JI!(WY8(Va&4!>J{NL(SxdU6BZ
zL3#0+Z1k559+3#VBvlPQksz1`8)FGPu)h|&=h1K$^E`1Fd4M_s>bhf7$|z50zrSmG
z8_z*@qvCW2^$WyI|4Kd>``q_TJ!Ko9#@J+JhP9Pqy()8CohB7Y_rX)EU#r6WB(vDt
z+68u!wt*g`eX@6fccgo8zT8zh6dDaBWjZwmpG9vo3-UZLMVbLB+7Ecn0+5qIcaH4`
z%LqWvbB`xxdd`OvP$5!r0`c@~-y!-|@HlMYUe3CyjYSr{w7SH8&`X9#Xas7&4WtIj
zQD#yuELa7~5l=1SZBH#ls27-s9*JdL?Z9%doWHBaq&cZ8+C*`kwt!9)KZ(c=$?kG2
zQ-jXToC^NN|0?Y6S-0Z`xM*+Vx@p#|UQgE@v5w!YDePSyNDHy{=$NNp_J2}&LxR==
zt4a`zA?XK<B63)(rAxv}bW7f0Zz}V_Y`;f!;<?t9WFj#>v`-!i&IdaN_pl278Z>jQ
zP`)S!$&*|1#7({jxCHKnD+@i%&xp>Mt3z#F8|bZahv4hLIH#<=b&sQK;*Ise+G@u!
zw3isMy;z&*cI+~fU%4toH6@eU!w<uz%TH}*q2ROY`_qlESn5plC-%Clt3%|x>_Z@p
z+7+^>&2sjcbCn<HdB6+r*6&iMDOiddZ6h2_17qkL%vM<drKn2O4RIpx!`grW_Bp&U
z!lNsaisiRZ6Ri;2K$b?!@eDZ!46UBxI*RUdV@<dC3F=+TH&Dkq-tdisVxs00XNbjg
zDz;BQLABLd;T^Gu!8z{pQfEU$yd3{fDdtvF^}!Kol;w)F(o)kcSPBTyUR}ohC)nNC
zDQzwns=+H~ZGWP+*&pc?Fek4W9HLc%!}uduoH#JQvz*IqVkPpPd|*qxL<4<IQ@~$=
zbhU(fSZb|)#q(7^YN1Jf1G5{eQnU^_@l2n~@i5m)y;CBUQM+h5L+sK2k&a{I^m$ni
zt$o3L_*A^AX2C2e8s@uK+q*iA^g+7Z&NkFAVkle9s!`KI6+vQnckp*mO!cRu=sVsY
z&Nh}J=CJjKbq2lzD?$B(FWtO7xXG{#+)_#OfV}KGO8*xaApPSh0qXHDwITR&`l5K3
z?MOC2w{3g<4S-ckmFQqOauwBIcuL))lF)K@f4s8o0lmb&F1J|wX-^dfWFM!Upss&F
z;5;0_mcZ9Kb}gG7sg?auF#Atu1u>34L9e7LbHDpzuzPqJn~yw5f1uK-lcpweYkDv0
z=<Jm+2AeJS6nx3Kq!o&>r-imKI-?G<!L>wj&K#IxOAz;~sg^eC623N`=PfTv*bw(g
z#~Q(%Th_l@=qo%`f8b{fW0XSi8}*9%sLZm|#Mf_mq@H3XN~6X8aGms14Y*^mamprC
zKcLGm6dgZ<var&qJg{J^)xXr%l#6TN6Ug@7k8(E=+u7Ft$@<uOS|ITpj1yCtFf~ZH
zLA|QIkEm-sFBLLl@kvyz=pTGL8)-L6Uc)`f#gEgng&lGe{{ylW@hUGD-y~MacH4_r
ztdfRfjZjtkn3aIf6stU!t7q^L_o+gJJTvI;mTF=;XUtgUsU%b&_XJ7@Us=9jZBR<!
z4mfEZs6Honw3X@x|KSgr%F}VuPjNo~JIJ+wxLv_3$~ey)6hoAfmMUxF|0V3k46!6g
zhR@~y0*$eEN?52Oj8B|G`QvNAfV_|_$n9=%vo^gR>|@EXy{7<Zn;)c`q0#a-rK+zZ
zMv_mmnk%AQmiUy`G3$ePp4^Pa;a9D5xun<_>N06kZh;5!K;G+}sMX`Af+O}m)Sq-I
z^De-vSG0h;j^I_cn4s7U-sWe4=5j}ETl`G4MvqoI_^ZNCC>t!0`qSst{#dcNP=1s+
zLYm9(p{7c?C`((Vn6W>&m-c3sSNJ(`COtzr2R@32nR@Kp_y)d#wuf?4x{s1(orOj-
ztMgjxBx(Y_?%Wh#pB(MIAr`~;UeVY?n2rARH51y?Ljuj+U*)nmsiZ0e#M_eJ=wVb>
z>KWY({OwI8en>%byS9kL!6Iu3zm@tog7_8eTxFG^&>mLvf?dhC#%we@a5CYw?T5J+
zbwq&DXY5~Lh9iwC!_DKeq&7;CeI!hhPqW9VpSD5ZC7OV~s~0GW3K5AQhMsHd#I*IU
zu=2RX=YlbgdP*^~6Tb;&=dSV$vVN9N;t7O<I74;BW(n7bJHmYR138TPB!39r^vyTc
z*8jjV<O*msRRUW=y@?+Z5&9GSmPCx00FDwQ-o^IC-^*vRw-koC&j+4^`5+2ZWIAb2
zV3fO(VnnluRzgkiEYKNC4_rY$VKC328h(%M6m^x;#Fq*YsFg6^+K<{4tY>*3l%-lL
zR_iVI9;q)`!)(C*V}{%7^EZf0VX=D)HiR4?jbq}Jmedd<<|xEgsUxL_STSW(a<U^H
z6IGcyAbz&Yr9SwPtrr=sSNBh(vU#`K2vdkN{AGS5SvqH<)fv@By+h~9t(hwNFcim^
zgZKFQ+ZMB>fYXGi?eul~xrjo#sQynb1?!_F-f@8+)+zioVm!q$2E3tmJNt^y;F`k6
zCVf{2>oXmF?emOt9Fv1D)c#aD9J|BBpY=TE`x$nL148}Scsex$1pC8kaymDM%!f}5
zZ>2p-2HF<x>U^Xo>80QU_N_L~N&qr2JT%@{8Smp-E;Z99$_3;_?1o~~X4S_)X|b|@
zqw5xbBma8sE~XFkMtpLlZZ2ZCM9nN1N~iOcLcO&k?&|hAp)0H|+(P5ncSNDHJ-Cj&
z27lBlM&XSKplWyq8YF*}g5pENFUNbm6r}|pU_Tid1RT>$??DA>gR?0*E0sd82}}42
ziuh{ID%@@z*PAOF@(=CIAlLA!)FJVlUXS?XIG=Q1s$d^mu%_PaqE~Q^{TjL}7Y7Qo
zbmfh5hV1PE?5*5Ibq&sQY6;gmo-<Sm@32g0+=fWgTy=WOmz1a2_25g*6pjOv!eOeR
z1_?+vw(fzjpeXNeY#8$!J_D@gd*Nq5H@MdFyzU#<ve0tX0j+!c<R2^;OX<u(y^5n#
zjWOy|^)HNPy2(y?kIlJzJh{q{mcIrKrN*PvVm`B6FD6&`XuS$t>uBP7g&y(cLX+^+
zq&dR=NcT(yPM)G3*6{1qousfG)fX>_^McW6wEr94IzWd@D2c>*t08eLCBiDfHp&-a
z4SiJJpc&zEd1)|*7=*c*Q?A$QFvnk@yqvCn@ZPs|(Qd$()D)$cmI4Q1!+>H)l`_$E
z;iV&1E*pI(i>4PRwWaN*rc4?3&!PoFKWU!wjGrj=Wj|8~Jhj9z-T<3PNUjR`zvz)5
zEolxu%h)+kSy)cb(Mi~!I%<8O7i0G56F@Q7R;uV0xof^=oz3oG9>7+LU5#SbxGG6w
zl<InE)X`s8Gbv|k_?_8~ccGP>+i^u7Onk^4wF@TIqMe}%<QDLa&#;vY??Y?h>yXM0
zV9Gi6>aWuufHTTekR*2ncJM1FLz|`0WV&t%riDCk0sTlF4KlRz!Xvm+SS%(cl}5E0
z2LF^)!uHxwRxDJ$!CtOw)E$=9zaTtkVSXQ_7DTrFYz2Ii^4)*Ju}*0q{!YJhd{quO
zWE5n3Km|+Vepmlif5JxE93q7%(w!hd`-p81exbJ->F^Q0rPzmk<NjSAhyP8FB<BSW
zQr&~MbE>j`D%;iMq@hWx+0|$jHC-D^Wrt^mM~d~pF|M81)HNp@=DVmH_ocg%NDp5!
z>7X+&>7_auj}E`L%=G*Z6$cZQ0|XMfhpI6b<ydGJw!-K82LXfYDHaFTsoOvS8*p^g
z&rrMJDhD1c7AnDiwZ$~1ch9q99vE#LMy=J}`0o%Wz<7EnG=q2eg!-R}V)>4oBz=dM
z%_SUzoExDjFCjP%P6gc=0v*qPpdV2>xhe%PSSx&D+he5@^#|?J2dd*$Rw<43VYax=
zLso~L!`K-9e^f&F5PXH#4_?TRLO+XMvDM)9(1#>sX{33bHGM-u6X7Q<-!-AyeaATM
z1?bED63f7O#?JxC@k}ljYs8OXUU?juCt*>&0qkki(T>7i{+fPN+Y~G(tqeC5RPny^
z4ZDMV?>tY>)t!pVmFqZM$iqe6#()d71J$+PVN+MEW2(g#{>rAq()wKaW^yrj23pA8
zYKyjRu~rO`zTx^fdaJhCeKfQ}E)u$i9r`WG4GOjS^a%TL(O^Hz7ctw_Nl;{3;0x(-
zvd-qX!jwTgM|~kXs-@Ykbbq3Teh?PKE!N`*XF;se3+qNzgj1CS|Dn)oR0P|z=k&^O
zcHS(M8RPIApsmo!=Ceb&A-3M`A0Ucyf#$Auj+G!pb!I!@)to2TuC7{1!SEaVZ%AWB
zewo8hSEB2piNU@xHQ`B1xxg*|5y#28C-+RrnIYBo?yu#5li1hlSaol~2>3&2NH&EJ
z^owXLTqzx7OV%mJ{O+G2*3q`+U7`B!TF#c@i^Y%mIq0Qykg3gjvdW+<XgsJZ{NWm+
zW`QCxMy_M-AU7Z`geS72gw3vB{$11zXvR|1e{6j@HZ)aBa&}jvsJreq^fizLI3-12
zXf;@4sB`*$-Jwip$FipA5!%BB>y&xISiK!QjSVMuh1`xR^jWol{zD(YRJAT3y6#=c
zr-=37-}(e+t6*vD7~WKC2AZ;C_0o{oMra2E40coZYCD`;>eLRV0wa2k9#U&jd&jpu
z>2ZK&v@a+}A(UV4B6SMcMfu3Rmgfo`&~SM#ny1|Kx0VkvH?6OIR^h##Mjv<I4}Q}2
zD@^cjs|ZpEz^;RY#<Kr&ei!;u|6>m5_v=d1P-PNbot~zBp>LCTaR2thdVNKP6{!Ev
zVKpk!Vc$X_ct$qtU8#>Yuf;aQ=HYmyQBVmj7K5sX{ba3_-wI6%_JB)l?}EQH7xpMz
zDg0M(KQ+V;jj!~J)>_(YI@h-^lmf3Zdp5FgLx|5g#IEA<^){p~3|7X&dyonp_b&_0
zl9p>@n5)zq>sh&;?sUZ{{hXliK9s9}pmzthm?rbIsOL`uYaESWAN3#hFwvO*XrIPj
z2RiJ-Ei;%L>Gm!9EItO#BuSzxz73?(Q>YoBmlCI}u(VJbF~>CyB{BPjKLYzuRW0D!
zi7H|L%ifN@8M-8Il&5K@)mo?*tdV?LJ(t%5Owo@yduWZ=J=9gADcgo66kL0nvcz7Q
z6rI)C#rVIG2}}|Grx>qIcHdIUxeg_*4jW*KT#)~cCoNMPjk#zrjC$@k=_=AzI+yVw
zb`u?xZnKYrTWsUy^zaY8_~3)^J$eY;NS>|Xu5I?O=oY;~eo6J>-s>mW(_wyRpZs{o
zH|VBHS$2ST^3+hnq?*oR`zPoQeq>#{%E>07yZpmh2+wBJB`LKE@ge7=bF}TPoD($U
z-(tf#Mr~W&@#=SSNoXDOjr)^G%1sTLiL*dQ3$;<fYo<&1Px2AyDraHO^+z?H;9VIk
zshi`Zxr1dnmErrWJjNcl>N|eZyQ`hJXO`}ULcEl1G+mY%Q1FJH3F?RQ41KN9uA=a8
z;$Ym5f<pDT<a)V}<#cHrJ2FrvZ<q3&=|t3r6}Z>R9pA0|EML!Xa(Hm4JU<z<qppDj
zJlnHaxV3YI_Y!nzqq1MglbN!P!JrLn=9mtb1^wzq2GM5Cht?>O6ZpG}C0)sQNV$jY
zXWL^9!YAtn!nLU{(9YoYub!HEYu9*Xw;ITQ?m5a86W`r6Lr0yS{8ic<-KLS&92C$Z
z%mnObqmsMOo!NBqBmY=FB6IW%A*B9m&J#8$<MGP4lBD8C!mIfSwyxS|kRNCQCQ@7T
zrzstSGvFmSF)~Z~*EUQ3o1V{<a+Gvc<!-YZ=ttW2FiwXO$^o<7fa?_e3Kxl25m4jc
zXkbzjlv}QBPNC{bZ`lRfZsnoTTCh=HBtKxzS{~*M!ph4J9d>;$JR@~+4%CLCWW}Y;
zqYhiwIz8c1uo9N9dDLd1kB0I3O?)w$@4UtrQqeHXltRs-TL|azi`pV~jOlay$Ow*B
zvm<;N)H1OWGdjGI_)p)bk0sNz{^2=VCh>`W9!LdarDaTiI$b*@PL}fwXHXRE4$q3E
zU}<NR^Q}0c)W1?2CN1+U$*I@b(Yf7%b(s(Ha&{Th-UYQ;^jQ9=v1`mGV?qQIeT1pZ
z(_lMw7fOIN6<Ho8ktm;}s6<~fe@1=c@&qNeC7cU3yDlVER@3}T3m2l*!2`zXuzP5b
zrB*msX-+iM=jQ*SvYaW2G5*uIP2S_GsLgYpP(3>x%B7^93<GD#HOW5w1r)1|*QYs_
z$RnuE+E(ufy>EUsR|Qa)>J9%O>g-z^c_Sn8_ETp-ivC2~f?c=op@^6iht=BC)qvRI
zOwC(fSjJIM*QPx{ZtW}okGzMAjeW}<(<aE7TrXF7T@G*O4~EZYePIoBAH5TICom-`
zNo*qA(e7y50twC~Y;(GaV`;bqY!UpZ6rgKHr`{`XlZt|8w0AP?+HP;8FQXe121Y=o
zwepdq!IIE%#{znekd3d9ALzfSB&bD{C5}5D6gcfo#enj`e@Y(hN^_f?&p~T2-JFb{
zXi!tylf96A%w=fjFxk>K5HdB_o#yMj1;(n;>h;iqFpn1orvM;N;fLuWD#6ZSXA|l6
zuFP;pcmA1cyfaAmprY8%j3a%B^h)YSIOS+{-Y$=KfukMwgD9I^4leeUcXXEOP$xCf
z|B_!2e5nk=uHxgk+2Ok|ma4(lpjz3!F{|W-_D%4TwvPUb_+-5%Pcyfn#>nlF!~QFI
zTW}RRk}vzR*jhPHwQlgG)|u^Ae^vDN0<*dVES7-tuN;_T4VT8v%5HiEI~5zw+)BEZ
zkmAe>bav8)aeOIqzeBXeX;1aW&NTEl{Wx@3Zn1Y6JR?q0>j;;EZ&A61TiJfD`CycT
z3mp}_j-|mL_JpkZ^es4<=@IOtEN5>y5Gv`2ac+l;i5W~R`=HUrI=AgJop<E~VJ~iC
zJIWgZv#EJDh)#e>RH6Bbwuq`%a2wmItW0Xg-qN~i+tof!O_@^=i@j4vsjFa1sIo|D
zf-(cyo6gA=ZD`tbe*-B)A5WEXyoC)Md(jDJWD={#=`q3i;&^={ImB24{fyh`I1rj0
zXOp5cUuHDX_oE$ZLj7iX8GRxx$}gm=xFd9o`7g-GH|+vlQR}5){sTKK@H3#3;=s<N
z)CPO`Zps$h-_*3=cy*FmQ&|m!#8J#a>VZ5F+@JxNisf)Rd?Xx*^z(JQDw`@&NAMoV
z=d2(3tuNIf`y#w54+*b@$B5?ntC>#umzj9lCfpC@vdhKH@So^OaEj<iisszN$bF=<
z)CrkoEyXGQU?Z$IZQVSWtL#ih3lbaFiB%nPKgDfdqSdhGD0{Uw79V)y{fKuD&xb0U
zlvM@34nJ@@{1U$fTPW8Lr`f;4*qoEpJa(hLA<~H_`iD3wYfN~JYg1^KJQ{717ec4f
zg*b?l!LEL<sW>p&AsNz*b#k+u7qzo_Leh2i9J5Zl4DSa&SgL~S$}iZ_JJ@iBVWpq`
zC0dhkno?INpFa|bY?@Y4P70RhJ2(@qrK|QLduSu-9E`b-J=0k2_n?1goM}{2Cv{>{
zvfsh{Wad*LttzNzZp^mk1$Za!9ZAE>_+RLrY7?x2Q=1+bq<wRDpkw8w{5bzSsWars
zhY}-h;^u0@9bLe4)GWeKO~e|t42{5D(>$Sz=Q9FQZ#f^dWu7Bin2Bf{=f`R#ch(NJ
zg5_wPAI@J&^hzrnn-x14>*$iONP8j2@*Snx)LZL9^jlz8UZ#`=(?Cn<6je`u7)+(&
z^sdTY>XH0E^Z8(d>@)u7*{n#Y=TLY1PsW7x242`->|&o}IpT;ohlZZ=T;3G_To(ve
z+$Y~tjgyAJXGU903U*ukg^p-ZT&lKF6v*GmVzCqP1z*nW<0UcKcSt<SUkdi7k-xs0
zg!9;5Wd(ev9&+!YW1V?Kiu3_XjJd~D+<BKiVGrW#?LVP|8H}3QMyty$i>)oPZ-tI|
zzHn7k-8Mx(Lyi>Mf}1;X{eHZP-X^f$bdP9iU&6K_W{~^rsTgl8Abxmi%Vpt9TDJGr
zhBMRbt5F9=^8Kv-fdjluG?cGNDd?)LrDiqq{Kx<WucBRA03_?*u|MT_>X_w(b&qqC
zR8rsSenTW|jbm%_56C4_sX&74mZt@m%dPY{I!RlzbDi*;QASq0c;~XnH1j<5Q0ymf
zwryaRqXWSsEaDfdEyO`6j@S>H1^|^4ddp1~l9fvYMxV?(6_<n(D(}OLh8pY{F-z2_
z-L%csUy1g-pr3=O%4%OzFh;v0M}b>-XTG~`2)^J-^AW{v7-{c`dgrZ_V>2xDV|q7r
z8s*><s@K;0iod`);utd(#pC<vCc$&m7JaL{F7c^p0X|IrWf$4S*i89p_DZRmeV)CX
z<qJOEe~z=OV_^@vkMJE#(#nFVWI1pc8_#wSU0Mf}N)G1ZsZ$iqcGo%Vm2ETk(ezGy
z=PTyS#u9oLdAEBAHrIYgyoBP>3%-Po`+mo}$@8`U93#0F+Ew=-Oq5y+pOM&{EmGE*
z1a+9dA-gIN@Nb5-94)+U;1sd1P)B`@tuDU|Rgw0yv-o%F0;!>}HKMa8qo({uy{Gqy
zP)cr3{qptDx~Q}CPX^NbH~RoP9H|mrX<YuyJCv@%c3V6#gzXtP6+z@`!C<ejvp>~Y
zN~Mn33&`E7&pOxKk=YimPUR?@E$h)ADrR~}=HXv~PhrJXT88_7RF?6G?6g^lVbNI{
z7IfAwudU{5d;UVpY)#lZd~ZTj?CytJcXch>&sN0Xc>mA`sWsSYGKwa?4Y`SB{Dd%!
z*yH9B({g^1*>H`h>EqyQlotqRE+R{T`l1<6H<l-}4L5}jd?TV7`A2+p+(jIZ<fV26
zI!nWtp44+$r6SlCF)|vSFrTzGOV_y>G^fuJl2G|NlR<x}g?Pa1fu3NF)+o@1pw+1C
zsS)+rQ62(b;s@XyDI>cjwQWa7643vX^OV<#3X`5QpT6i%QWj7?>n^+(?-mCee1WZU
zIq)BMiOjHoo#h>;%{B8l`Ks4Ix4<UbwsA3%#?^D*Cl*qJ$-yEpf5b6zd?1}~FHFR8
z{9VCvBF^)ec#L&S7|~!a^&8!oq2+Pb^;80RA?_)7LW~ye-ig#(8mV=`Jw972ugoAD
z*BOn9sH&h6(?=8e=bM(I?I5FKPx+wQkQ}EqwSHqBrHk0J03}0out7Z#3p&*EVn&!6
zxS6-r{>E6wu+ZiAuRt?^6V3OQ7aGgYKw?6zhyyYYg>e-$_Wy^?x6NmsGc~EL*l$3u
z@=NS5e5Dq#=jg?rww5FOFJhjlpKu7BP8`PWPnLvOOI_km@jotwKFN-tst{@RFH$ux
z>7K0>!6un6*Z-lmL@x-9Zm(P?AsC1owApa6tjf9KV&#(L;`~_O+Ie&(%_}VxaZef%
z9c&_*EV!ke*homhN*hzM*YcZ)9)T|I-dKO!Wt*Wiqwny=#9(2zy*br_jKhJj0d2Pp
zRa#iP>Yc$51mWL4^M-Q#Mfrb*q5Pio0bUjBOYh_RT0?uc#E-+LV|Q|2Dt&Fsv0(`Z
znF{fH(GaP<Weuic|JfF?wXp_yigmncs06&HiO%c>ccIuDC2Fd@DYX_m&V1CoaAMta
z)M&9bzlvCC-%Al7fVUv;pm_L-J|?f={M1-y3vey(6eyscN~>rhzC1C}{WmI&ol_Q(
znEKqfoxPB0MiYHaG^b%*-bSK+#L1uGIYe*w)eM^8ca#lvc?8{0uAz;`V~~}PB1?k%
z^nUS*@QMP`2dWc!4Z8ph;U3aTTJx1=+Gli?``{O}rM{0+h_a=hloRz@nRA3ejA-ek
z{)7Ln$2iMl_dqMVkvC)eJe3kw5h^`Zn~6Wi`f<A=d^b*R%+$oLkQcbOzAxf7>>{?z
z-^)5z^Q#Y-|9nHSe`>Z+F-sY$X5>WQ3@%&}QP3EvWRa<eG|CZ04!~PS@@WtGd@vci
zD=i7k*`1c#+xrc_pKq|-5Gq+GTBlM+v{auDuM6`0+p+=bD;}mRVABzn-hivWZ#;r%
zDaK+G1cLq!s;e!jc%oVMff`Fu1^AEN)HDm*XF5sE*a7*z%4Phr<|MbkU%ukNaIPqB
zr*Pa^*H?p~vFiFjb{g-(qm;UAoy;qS3I4rGSB5gLg*?_9`|7B#R2AD33$*D{`N&i)
zI`WT^?`X?H4<r3|le%6ufhu-8*4HshJ)sRD|FX}@KSf>K`yt*1E(MM;CGGD;Ctp?Q
zB{_0~)K0uD&wG8!>+M!_wCW=MUwdk9c_d;r<xgC@NQP=7m?W$f&%uSnZcC2XP8q-&
z>5cLc;xjkd=i|Ev`)2fX)JfbPe}ZokcxK8c3W9Uh_K-yq5u@$`3lw0vhlcsK8V;k5
zxzVt#;*suKPmvv%BUGxmmKq>-=Z0Hqk|o`}<j1&Znya-|mIW3WOYs&;=POc^rGbPm
ztwitxHl1M^k<YVyBWqg6gEH2~>PRqBjl*POy>+lWN$-Vq!Zvda10SjS{(49kqm;h<
zX8KLu6zf**GvAS!f@)fpNo7Gd+is95UIy2&_2f14W#1!lG&>c}ji|3)z8ho*%M<Dg
zf6S3%T}a<2OJ&<oXTK}s58ozFRbf<ZCE*wNYW;v8u|>x(U|+HOLMKHt^$TT6m8sFl
zMl=j7at7rJP1hG#J1S?HVrU}&p^M?zNR8;4`<@$zTI3$AaUrXuHiR{YkMrx8<M6wZ
z8rK_NBTay(h=$lr&`Eme?WfwPis1{!L6+*m5xzJ7r%=tB!p|kQipOkMf+?GNfK|E8
zI1V)A|1C_WG7|E+Gu*J~SMgivpUOUU{r09=nM7mc^!q`Lz%0rKjBqC!;hCz`4qiZw
zeSZoD_CTn$qT0#5e)K0@MK-J5!A^Mw^;4K6-y(;=J>+^gpFZh3Z9i1yNfw32nGVqH
zh{P<$_CD`DH^u#*_G$M#CJp>Dyrr)4m&i5rIpqM>N%8vPQ3Jj<aZ!Hdm$(L?D5oX$
zQdr<|s&xS1y4(9;E~YZEQ0(amsuj||+mFH<%5!Z#naRD;I{N%{$mvi5WGA_f@)TW1
z?znq!qu(#h=I;s1^e2|viU0b`Tdv}FF&eM3tE~DeaMe<VqfM>VHv1_ln;%NPP#zg*
zY<`66IqY9?Gu>;?F+EPGa&Ustg`Koa5F7Zh>`&4T&Ovun^4x30w%Wh=xWHszhOdfS
zAsDJ04n3?GpWcO9g>sd4U|5919T7MQlN%(=K)tM6kr`bL1gvq&OX_zdl24*`Q^%Mi
z;*t2F@`}8(RE8J`s9H(wg;tAZxNb0$RFwDTd&V(_o3-EJ>GI#4$7p~xv{vK>1JHJX
z@BXsde%p9(L&!+j#n(lfti$6D%XRcP?iPE3?%~^S*oD^??x5-TVEQ4mh8!WqY&6P;
zxjxiQ>6G<9!y|5QQDfT;ahlYb+-ctkUXf$H-$WYJ&(iQtrbehWH%@sdOfWa|&*ulB
z$@qI<q5CqUv?ByBUZUpvbCku_wcvfAklr2Pq4VS}d&Xdc6Z;V{kvyS9ca_V_ZOmtt
z8qAukk(e!#s9!(>d?!B&`~yGBx!$_f4(`=aku_J3!wlj&_>Xe~mI<G8b-*NWOKv2;
zAu}MTc@Yl84=J1Q(}uZ`+~*0{AGSkt=o$7MewqD?-yy$A=%GyZ4)p^ND|Z2Z<+1*O
z$ckozGlAoJ2PH}0j}7KK+Ag95;wD`o!r2;I{>DPsUg|1y(i|Y(pkKaL)Ln(5qOB#U
zLE(|LHe|NcIuPU1%7E2ootojcGUtsoeNiD@E!sT6*IKDa<a$SgWxgT!NwJhCNj~PQ
z2J?tm`efx|x)Y7_kbW+lkKG*DNZk>(V1ua7#@p8484X^st)w}4)xIISG0~4%L%7g#
zY^!B}{!pQ<_W}d3Q&=sT^i8#|7Rt*L=`6XU^iw-2NL0tH+Y$>tlZ(aaT3P!dYI|ZE
zM}pkj|G~JADl3{`CAk@O0-qnEedUFUzFFvw#pRzt{s|`M-BttWJ@-VOL!7nD<bHt+
zvLAj_o+7*G3I2=ZXrWzbqOb<992$$0a;%{|b0nY()$~AiYj6!5;iiP!o6>0$>K>WD
zo`B5)@6dkwRjoWF-f)%)pw-x8{iKj@O)^}z^s~HWrlZ$bL*~0&pUM$0SbKOEXvWBH
zL^A)?Hiwq5Kte2X>sMk#*wbIto*>>9V@z8cyks6kw9E-HiCIPet)2q4)Kq%D%u?Cz
zZPZkhXj^XCq-{pWY=@<a$`ju(hQc?K&IoP~<Q3r|J{6Di43+=3x%I~M6*O3VWbA6~
ztxc0#^X9y4=^5W0tN;pGC{?wWCwrpdwrshC<u}B76e<~Zv$cd1;U&4)@Y!||e8Kvd
zN35>^R?rWLrC_dj()0tIBR7%Tuqe5%QrX&Hc#$=N&W@yPE<u&R!2qn3wpOZ==b+RE
z-LjAK+xf2a9#lcw;je`167JyI;COOB^*{VT?nBESZ)SWGdm+=spR7#sy@ZF9j#4|h
zzP?#rMg1cz54@5;QCHoMq++;E+^eQo8-c6Dyz1l3iFG?l-DSqU2<{S2;D>C#tJSj8
zvBMN3%hJ8`YVlFROnxnH2jA-B+Fa)E_{#qM!hRS=H^5j!Y5pb8qCUAw+#Bp@_@QzM
z464=|j^I;*2Bn_(QtEH_I`XM}tUYL^WhnzIWV!ABVJ0DP)%MPEmHZ`@;Tq|3!Vs!C
z@Mev5&c(m^#%7Pl4>8f?RAR9|9jwQ3ri}25pYJbEFSKtp%nQ82%L#`;HunbIw705g
zB8Ka(z(&CxxCb`wox?rL?r%&4zEE>|F4tWdm8Ytu@K?fES3Ui%^uc>hpN0qI1oLWg
z3)aWJ+|Y`zo!wNxsqykJj~mT|-^t2YCt*2A#Oo4_x3ujya9OMd1}WM6)ZB9<1qWlg
z`-k@s)0G|L*p4;EuG^BhZKx6`k8kpB16$H!n6_DsgKPMD{$%-q-A{jjn=_x}Rss&Z
zzIB;!)_>OSwHAOYmgjk!4GMPM+f7}QamCZzzFWSh|IeM6!);>dmwZF-W9C?9yWo4Q
zYPyeT<=vey)A7r^OxYG(4OaOE3rEFie5-_pa$8N*R^pAoUTz+_*7i9+BYG=WI`$!&
z3pyDZs7p`=TuxZhSL>uD)3i|DY;h8wh;GtjbcR<FPRjR*4q=xe-_{1T(x$<W%(&3q
zLa%liT_>l=-^DS81&RB#iEw`GK4@0U2Nb2Xau9Wp&jqf6#e6MPAYG3zC95(-9E@Ke
z|B5J$C+u2g?2hqZH#%-vx`QT=Hb{z&uc7p(6FlXu1EB#dWk*MtO>87hzf5{Zt&e0)
zB<2N`YHx{rhAhHk?gUx@qqY|Rfi=}e=w+>9Bq_CkfLy%MrX18<Fv-(eg0;sB8IDrq
zSNXIp-|{ukQvFs`2R%jW`QA)*@EG6in{MhB7v4dz1!7x%mO0gzk=2N@l7rOsE1iTw
z?+(9WP9&DKre(gz!%7Z4lx!}JgBVd=E`#^va=mSsB5o-jFP@5X(WB)3=mWjPG7_JK
zO^qK&?Wz6B)0o<hg2Ym}TvC+U%N`<*YsUgM!#d?Fyy01$SArhxScs<ikDB_%A6GVG
zm&H=35?<H0Tb03LtR1$5J+5Cjq(wG*UZUR&UVReX&htB-j&~&YC?l!CXq87M4)~if
z4!s%WqD0zc{UH9&ww0@fU(I94>hutvqSH_hG=!fZ&gE)~HIzyqA@>;yS(kyDloL4k
z5r&P*cWX8o6Mr$HcNa4$=>zy4G~t_(*8#Ukm0@{4oh}y23Vjk2>fYhMOWS30;2-|9
zxs$ao&ZA5dFFCP2=s){^M22_(k0zU9FVGvV3;1E)$~VFm$D7Tos0IECcr&RIF-3lA
zzX=|oThu@%&KlN!1fqOEpfQ(cA-`8xDC+Jmw9$9f@I7I_$?tvPzQeiXCH#8s4z`_m
zAX<Hs@p;5-e2Tb9yk4_Ac0d{rtU`-Gb7iA@iFK1<Yr-@#7Q9Kno@f&45hinx--P!j
zbIf1-+32&oOJ1eCho<_%T2JfPmVrBB9g_pC#h2IzdPif)<RAP{ya?USy}iAq@3uAF
z$HzqT9nb=KCfJHktkR~+orv$_tV#z8b4B$Af7RGWs6*x}zkFY?=X@n`5i!N{G4mY1
z(wF7ClpBYg#m`G-lPj`kGzE*d+p2CBj&p;#Q5kW8vRSdFY<QvOO1=_G@QgE`jU>V+
zfIoe;h-dgjFdZ@pW&8(y5sw?MZ+XMFww>o1P{3bLX(LbYLZJ9k-7JU1OTq--Szius
zuw22W5$nAJglFj#=<mLjHZCC*_sXpBEwIhf4}K$34gZN}V|I(jrD1X;Q;UpzR(2i#
zHsdHf!p}UDtY7jjAtZ+bN!B&S_hbdWF?GUjME5KhRwa^2*+PQ^f_%=8;9JF%rCR0<
z78m1FjZ?)=wsG#+{+5;__$axDa0O2=Kgo;MNbrO6;>|1{z=1%5Sd?HPI!iCS`-wz*
z3-cy$MJY{87RvF>g*9lC`IpHcmlof6KZ!EGk^d_z5w$Y?Ko9s>yU}u;$P!y7F5&N`
zPmJ_+pEMPJ%U!YxV5Hq`y-tqg-^#3^nx%rRteqx*qj+0K?qJ+1;fjs)m8Xk+2XiK(
zZbGVOre$8a-nNWHU&J$@`Ao|q<89#!_Jk+-_Cy^1m9u1};gvufmg4>Cxkn5)KE>|_
zRB5G9BKBkr79T?Zbd88)j01Cd%wsAy+Uw>TxvPWu*=ywr-VE8BcMoeP<ccc-TR?Np
z<UJZ+ApEGEBDV#k_ZPXk@^$mXY=SuI>kwGw{@eZ!zlEx!)e<wM5Y1)0<R^g#JLiCZ
zy|G*a@&ezEI?ngyzuN6m1H7fYj(;b{MLNZa#2@|{=tY_fn`-$^#w5HVI|!vofj>rG
z6Rrvy?0bmTfvL7Rz8bQGrdh`Ot^70F5#oi=T^S?fP%BK&`9JX1%J7{-xvK6`#)=*{
zxF<RBKLRyOzem)6l6c5=QXZLA-S^A86u(MTMVqCYSseuuO)#(aX4?N3n7FMm7qb5d
ztg#02re&-&ToVT2l>E$cDEouoZ>dJaOO*0I>wRIpziZB{*uBQryzcDgx}@F6UmQ-S
z@^zXm4F>3&ygA`v!65H&RP}P&Zw2xB7ku7e6<NkwX!V16`JIFF;BQ<W-4)yq7Q)uB
zN~oLfhpUe((Rn<S9qN?m4Bc>kb`B34-Q|cV#o~?G5yj>tS5fneE~5$Xta{e5moDe8
zgL)=(PdcMyFy(gjPb%x`TrkJ59HxYO*`4OzaF(kQ8UxE{$n#JBntF}3SG+gBr?^br
z7Mv`14Oz@@lRD%N(dN*8%;k;>El!RF<@LMKyb@#CuhetZ0jpSh(jL7YxWS@=?rvv(
zAL<w7%DM#>!Qp(0I#6B*M}&v52Wu22jjQuW9IHHp>B$4*2ZeK(McNpLUmlhWj8h`J
zOH(1moP(Q_qa)jnZJcG2QsFD94U8tR<Q|6K&8l-^Qg(2q_?(Uc?~?Oen;h%lUe^I9
z<i<N2%R`|Po13&0UfS+1_^9VfGk0B4t>K+af~AG)fb1?>9~!lzKf5Ysci|t&uf;FS
zb;l_-Gr6>DuVuKaLQyALa=NwZ$w}C9P&?^t;a=rcftXy3C?{m^J6=R`Dn07lf!4aM
z<*ssXFo2Fv&T~HVYy^2>$(0Uqy{BW4Z7uW7ao>x3UHN(J!MK&-*Kk|X8}drwI5n!?
z@}$d=eXDNyF~VQYRnV2sfHIM)&WGD2&B}bgtEB6!GLC&5@f|wLlk<myVE%%F*ZKFH
zitib<vtUEvEM*@&1ji*Ce2r0*z6#{-y3KW!+Gu>Rw7dtqth|&;huLh}l?O(GF=neW
z%XJslqvt2rMVIBCX?xYPyA-;jFH77)c1r3d9Y$qgwXiEcTkI)hax)`&`;Crxcn0<b
ztLhw88lW-SuiBfm2hJ(9)v-N1P;_XC>I}zPCRTnC**56pIKrf13xyM*9*)!0Hm%6I
zHdxJWkiWCn!qK76!E0&~<7cPnER^Xo;eFPoyM7C=(T~D`1#?ve?hjYkHPQ7MY*FSG
zOn2-PAG39nw<eurpYE7d_(^%Ga_p&ssRSi`XI*qX^vQ}q#h_oI$j17?INs3%p3-*6
zxr(EpMnSIcOF|}FD(RCWhWT!&PR6iay$^d=UdHla#6tZIT^qh}MB$xDE*M=f%Gohk
zBm9salK&-G*^vZZ(;I0vyc#RJUBmtj^-S*W{OVklY*k7W+{?ef4)HCLet>Pkm-&C~
zJ}q}p+V0B8ufl|IANyTx5ZQU19PzM^7t;9$T@|!PkvVFrMmb8e&S0tVcMNmQcfHgu
z1jnk^6uZs9*7Tg9O?b4rn{Tuo&wpVc!iBnrr=26rZI!1)8_>7#l2*&PG8D}mLLg;#
zFvq6Itn|oNE!e3b4W0@`V>O&7@XYWMiVO?Hf2azZ$o5x$qaTqilT+z|u0z4;t_E^^
z=z={fd5NQ>R-XMm|Fm;?{!Yi9@J4Nx<Nqi+3;(p$w-3)wr`_r;#gY&pfg}V<=kD&>
zvz@uSyL4{dRxcDuAV7d51Zd~Jot^!*Gk14)_p_JxKgcJaJooeDxxd$S7lGZkoQtif
z7L_$HmJ`Vb(H!-nU{Rov@rwQmr3uW!{}nr77ormXQe&dAE;=;y4quB;;M$v1)N!dh
zkSUmn<VXrIBwUJC!#in3@(1=iP=a;Ee&ZVjSB0Z;JTVuM2mjI5^M`~up}XsF;GtoX
zXBGLA{KFuITnH}g&-%_2q5~NdY3=D6X%R7!FNm$~bKKbUnu1q*nY|d4sxs)7T#~L^
zXff{**RqMoGUGVzfN+**73(5j4YkT@%p79BuuPyL0GfKK8$@*Axllr=4zbK#S!6^e
zadX8rzFq;T<v6NfYmk%iHsVs&i8jVA;LnV|eOr+tnG4Sprm`AO_i%rTlRpa8G43J<
z0iZr?or`yfRF=Od=JKtHZ1E)ClT`70)Q>zg+zpxKuge`GQM@YNGN>~S(ex!>7)n%U
zwX*`p#IyW#-PG`QTdnna-#KCen?rR)H-$&wPa<=WT+@Iom%2TM^2*=}6z1oVn*uY`
zt#nQK3x+!cNGvx0e|+q^IoV)Ue8DF1jko^=4-%#3ZNv=y$xspBK9q;=CcB_b#8ks+
zKp(on*2af%pTLFW3v@(gwy{toqf>pG`JQ1QT!k3QPd7})YwA>y1VqFylJcM(Iho!#
zvY#vFS7xo&tqmoFcZmM%=HW5eQSm&nG;%O96#Jm{6LV8e>%y_`X>{nbcMLKJ3yI0*
zF2?cUb;d05ob#(e%ij^?!jOs@Tn*7ruFKf(`HgFJcB2caipE)1>(1N82UYkMLoMdE
zVYzw<F(cv<E(k-!ot0tTP-3L>PG~2V7wKjyf`1VkWB!Zn>S9g-Js@v!t&M8sEgnZI
z{TpIrh0RqjawXW5eWm|Nc2oB;o;J<GGKBd==WuH=Em9M241Xc)Vo$xyyayl4&o;f_
zinzn_?wYotWY^E_b&avKiO>&Qr=Drv8#MawaRamyJs#t!z#>EKa5uEJYdA;gb(kuY
z2KLn52-YTtumU>@{aaUsEFcybk|J@c9g+9OkD04XS4qGUCYp0Q@Y-m)VV$9^uA$Fo
zx{p?bt8vcg`sgY5d-fKCiQfa?kkX6};o0CIW25k-sFQ7Fm>Q}Tp-89rgeeKDyswP!
zjLZ2p=m4~*>5dD>mBetWmC20ZCSEs%T#tX!pAx&0``M33H!(w;TBb2JhbDvfg3^#Q
z80X0_VCV+XsQ*{|CO=Y9-*C$~hqxc6D<%+wu<fD**(RUk$!)R@uORZ^m%5+gCelj?
zdS~ne>krwY2jK0odU1{T#WLcsitW*9Dl0Oc?1L?|zKkW_8U<^463D9ZJuIzP7~T>_
zcoeoDIa*;+{uN8#d=@%zUwsp>qhzv<jB1ItSRlCBKizLBjf56vzra@r7ea%8Tz-jI
zD#nv#kq>BB<afHO;%H=wznjoeRYB&Wm|?#Aqo)$pm^*rx_&>2P$+*Dj=pBDLr;k;b
zXUIwV4*MTZa)byf(u@o6zJcv(zv-v(2OQ*7!o_fUj9S*95pf{mLsKGK2`}+Sb}y_8
z-#BWmNY)$+Z?4qq6C!yr-^4#&RO~{YGOWamW;i?<kKjx3Ut$71hR+qUOmdS|+MKJS
z->0uiTn^qh98upQ3zXT8gW9TDKS;h_HMox-;yy+85`IHh@)e6{`jcsJG^pm1c@e*q
z@t-!$KQoj<<PvA1r}-RzZh1C#(l{H;Wozn}!4Ld%NS#?5>XX$D)DkAoD0H#76@M?x
zCo$B{z4c`a*R?~fO3!?}9oZ+M2(>ikqH@8lpA4NAW^3=W)0lCA-TV^Hr9SF2a|;9W
z3Q|Ie8Mi~Dv4ufnqz!hKsOtT}%`~(UnCJlHB$lkt2=|JP(O3Fc@`p2O_@@bt#m1UC
zg0kWyml63>oJ%G$tJSB8f3e0D!?-1;srsj4HSsLRsG6fje3+>?d1c_DxIGZ#U{qI#
z>4sy<U+icyHBumQhKRcdOXEi>{tY)&e-o0_Yq{##*?hi#X80WXKYx98R{radv2upB
z4!_l>Mt+8JgejGOg~waFVO^OG*r`A`J;_HJkCA5MzkH(Di|ZntQ@ViPL9XKuxPAy*
zx$E#`_M#9+o-xhgW(xg?gYb>uCRorP6jm5!aXZ5+)KkO|dey7sP9fVO`{|D8Ufyd|
z`tQYf7mubmkSDhG)m692{ATLvOEcaGs8~HeK)i_lFP33XqbGPjhmzQiw!ebikQD86
z^E>jHu0^<~5P^$At&M$x{n>ha4&8;L$(rsjs7*Xax{7-Q{tF%t<&%Rv!;#5KFPF{s
z)>l>I<_gnU?<4$<&{G*Od4&debCt>15PuUY_P6)d_eeE&y+?H2pw)ho&mzz8Vch_%
zzyGJ<B-@E>Ana>!#IQbeM^uC+viry`L<>W)cT{-?lbjys&E`x(Ix(2<&Q%HKW;snt
z+?@4BIZywEc*gzBYlCxy<;K6m&xk64<Ul4N(;pG{u@8jXXjf}dXhQTq;S%wM6=g@t
zm+L$Gh#0H;U7aHP8XRewT~UBc58fyprOkK7IIzHjiYw7SM1m=mS}R0AjSC78MK#;W
zF2-1&i;Hy(0?UnU5z$b<{h~i`Mf#OGU9bVUnal~fjRImrx*Kn3{xe(%dweypHb4L=
zHyp{ziL{RFSJ(HS1$^>z<YIOiek-y##>TYPUs9Atb`fp-1KktspH*9gvDi-Ggsi{u
zzG1aE%%CxWro&9S_A}lnR84nDlgzuxX@Gubw(&1{Fa1~X2Hx8@QRPLQs@ujKe3|2a
z>>l<nJPECC3|6RugTezCHjwT$ngfA{o(bqRtaHSrR>hKRySOC(_3&B(jkZvqilsU=
zF@k)y12FY2dy88I68m55_AW%mEDV^&kLc?dUidSh9#JTAOsUjQ#s@_Rtfheqr}-0+
zg`6@RhZea@bsWEvtL00=SKt%*WX7h=*HG&Bibh}~(-?m<)Ib)Y+hQ!oyx7`ORc_SE
z{SC-E7%KMARYy^xnTLw}=RO<w3K+69SrV*Fx1C&#pV4myadI#E9slH0qd(yf{Av@e
zpJuKjGzwd!O>vYw6}7V)vNOnLG^&9sB5V_)YibWP)?+1m8ZN60<ZuwypFwx>q_2Qm
zXZjUHD_Rqr3&cwBfquI&Q`|!~jLswnbMO33%>%@=a6JAz^i&)#ndq$)Qwvi4hY+2(
zU$m2RpvPh`&=23H>5OSm2hyCWscyoeS^J{~`akgnna9+P-MD=*78#Q`8M!v{LN^@?
zaAS=kglHE>hl(Azt2izGT`oud2pNKD!T#D(^Hfa`>p-1krv#U3-U%#nBr9cYE4CGi
z@O{YP>FZHie8t+4n!#PZdNGPuDdyn$*cb3K-v&QID)h4hE3=>TBcfxA)WTGJP-u6>
z4Lr|uQGATg@HR(}5Jkd8oaS0VOVMM%c~7#oZzux4^fx9ui_?X3F`L5!yi24^IF4Tk
zpAVlgZC9@iHlnhz626nsfs~O!bwl+V?-BGO_d4dS_&`3bC?z%pSLs{e*<=erXPS(8
z=y$<;{8uc(-NeU|Nq$?f((oUYgr3)D>K8CeRR-T%Vk@-Umka*Sn84IDT_TSQeaNF)
zEMP;c8F(>G*vn_Sa}4v1|6ql|GA=hdK6{txCw@QJ1JU^2g!=`GgVO^=#s&K2NEy45
z6vd-HW$;$`lJ1rsC*MPy4b1$HA1llc%MC3eFI_2E1ni3XjmP!NV<{G!iVZ#CXW&}@
z8GHi^d3$TzD9CjcDv;&A7EDuNlkNjbvDcIfOetv#{gIH;e9oTEtaO|-yeA%*th^;K
zD0QyTkViZ_!{ZgFg;>8oaEQPH%l+NeZ(VVQ#<p!98Cxb+r9PAEur7g@TrJMbj~91x
zHG*{!Cj?5qAQza`{Annjzl-$4s%eW2?XdG9J3d+5nF({l46RdY#2j7^V?Fw|q1sqG
z{sodsv|(X_wZdF0F+P(d?@`*o2isUaksrjCF&q8;(WZ7A9f$At_tSpSKSfq>ofKig
z7*Rm3U@OlCosI8H+(h&6b(l)n!}Ld!;di_>@Tc+>2b=yE+)9+|&_G*UO&%3Go9_6V
z@mmAixycbDvI~|NTZbjcTW@#DsTEvoB!@f+{1@1Qw{{kUUov}`k$9%@tvV;v!+S}Y
zV5|?HXB%)|^zX3a*fnN@sgp4UJgm)g<psClx$Fkx1?76hJ&lC#!ghj{#CN=#e2xAE
zn<}jEA2JPQ*KxPGmO&tJi?14(7+i?3-dE_Sz*V89SR{`0_t!OtIce9>bIol=&u<9m
zLIE@iSK*Iv`@>p66tkge!bhYAdNcS|M_~WMJ95pvgZL*716D}BBb(v*{%YV0*=zM+
zB*O0xd?%wI9vl_W6E(R0<cQKK#!p-y3_%{VwU`v|;qY!mEC1UyAHRj}Bi9iPi9P%-
zEYTeluChl0|9Ed0vO{s6l|BtCpfdWUXoW{nBeiA1JN~6$BmOq@0A}(7>NVhdqh9Wz
z!5d2jS7qe7<z#2a3dD<@L~D~4{5j$lijiL!;~T2G4i;lKu{-!0Fg3g)6a_WJ1DrG5
zkNH4Aasnq6NAmrdVf=hiBS`TD_%TKbKLg(KuY}{FzPL{R8ySauC{NYi!@?Rhag)2Q
zyQ*8u_VH2Pa#5>YgrsSj3Np{%@)Z9S?}?a`X(ZNBMq{_o7z>RpgbjS6!K>xT39Jpi
z#KZ^NY?*J2(-r6+gGrgY1v2x0g%*myFBogG*{WtV5Wd2-M_<J5$mSCd-EGCRU>E+D
z-&{lxdt#@8Jn<e16T_VgxFlb$yC?Qni~)y)a^?qB>aVKFAsXo&#6HI8NDPKzd8k^(
z&Bl}IPedZHmb>KFVAnFcAO=l$(@srlsMs*uKZiZ5USl9k&oqgF2SOkIfo+Jt3D7`W
z%TSJO2}~9orcK&}id^J^^lRQ{w}+|+_W-8mRoFD8G<49>*)=$Mx|}Tj)0UpT2Hu)v
zr7qI_=xXYwiX7MWOp!StF)}8#NvWVaG21M4nF@~x^<-Y_TPlq1pyH-?zHDbvA+VB>
z%N~^-D?gM-QSVES$R{bU`FhZ!(-uiqz(uO@uKyZb$QwwHSEp>NTR4EJ4)}nmqlR-Y
z9nFI*A5~JYFJO^xE}ky!2A)tk(%J*f>j&lMGwK4*N`CoLN?L=m;-$<Az^v+2)}b^5
zC<V+(7*GI=ES1}?zzHCyo~0rj+bwm-&rTL<M^yuE7u>hj%;@Y`ogz@LXt#4UCC&d+
z{Z6%m%2DJZ`@pBrAM7w{gN919(dj8K>=z*o^-zAI{sd-k${I;n))JmW?*>cV54~^P
zolI?<h0e}cUF8W)f+JBwFk0wA$;{Y|ZILbNw4@DF{FKe{I_sU&7{Tj6j=ey6&)K82
zt?!)b7(X3MF27RPRepq9Yy0JV2cO>7N_EG%b!$Uv8Q_Pr6^ruU!XEghU+LPVkU48O
z;_bsMAKc%V(ZB=mQ))MO$6l+XmAWW%3iw(++cko?uh|aH$PdakOB>P^wtcY<Ln4q$
z4NhO-93Sg)Tn3&<>o99Pcj1A+buhKO%)O{Y$-JhUTk}<uz{Gkril@3B%Bz;|p#D|4
z^Crg?GcbI~8<IRw4NR%y7M#_nq>O`17AW^7+V0TpbaCLI)Zd#vxysyg!8U+S-7MiO
zw8tuEu!{H4CvcrbN)J~Yg4`KZe5Q4|@~^hZwj^hBcY?CB><w9_z6?~l({1-HiBusy
z(>0;6Br`8#lF#h@s<;4jgKBxoyp7Uo!Fye~nLdd`l?prtiWCVMf4BtMSqNvoQQeb{
zz$>IPZIDwhKdZ86HUs^HL#n^*^_|`9|0thRD>C*$v&uFq^ll0K$uUfwLS@@~Foj@C
z)m7VEcTMHM(slAUSxcz5Ye~TzI6+q5(at@}+RQyR19$f=yKT?&9!}0y^^~`FSefC@
z&+;S6hORit8K#e8MCRZ2X0lnnlFU`kWschPL3vQo%`pmY?Mw%IIH!T*?Wh$|{pnJv
zwt_C;aN#pqW7iBBu3Qg%vDUUV_YAOo(sTg!K)d0S@Kjq*aGVWNes56EYDnKuKF!x2
zo(*4f97%a24=X1qyE)3rp2<v(hrm!c1DxzU=NwRS-XT*(3lBN(d&KaM^}iHX!K<;F
zbQ``PTWfu+x|ULxDt8jlLdCW6OUwX=!z$YG@~bL@Sr+W%kgA82Hw1@f%v0X7_w*>$
z&N?+++tNlegxaOPA%6_8^kJ8sb{GAYJg5OD--fzn>~S@r9{^udOf;jJQ<4mxus)!g
zs5&?YK`)^+dsKO*%m58`T~*h1XjPkCJTPCeP$oG3P%iL}vAr*Sk$6}(wLH&?WF7_U
zY^tqlEUWK+EE`*TP%$2^3-q=0wx>%I-Dz;kisOJ=F<nN3GGAxs2x&uEu6B-cMJWTd
za^%pgdkN&Gn<)<2&)Zgl&m3m(i4&BK0^*t6jW>aiw==LDx&!<IZaXkY<`_XQj-Tme
z^Ws2=Gds4a(;O$M8}tC=AamU<XJafpJf99ig^ntk>5Ls7q6V2`?#6&0a#_zx>)DDV
z!<4gNAmg>nP~>(#lKH{)Pzy<);tUxCnLIn)(^4<efNF|NOK-`XDeGU9=`zK#oyV09
zRXtl%7w9OoPF7zo8Af-IU6H<ko&oK^r8bgzu8FFay4sh3>KfE4N4}-lGtDY@9d#uF
z-B)Ewr$`?qexOfE$`n(q?<7~{8x&b(*MUaB&iDgBP=3fYg+1WvsUAc%ppDRaXHG(G
z#Xb0@%_#$*%q{nAljJVTaY;jGDt!gYNExa<NEH+n+PB77v4Qf!jE$0()(+~b@P5bN
z{+z&k>aDYg*=BX7=KzDC>cEJEYe@38T>Aj9b_UC&Xu7hmq*zhDavJ)A+GM{^ZlUw(
zt?7+iQmsSff|_Y+s3y?^Qrg?vxqodjIf|&o30}o3suTny9anX-q@@<ctdCjhxyn`v
z>*xzx2)3hhJ(Hq%3k7Mn`+dR(Z`jkv@&wvoMd5?A3-1OlN_eI^hE!X>eCs#nhh`nK
z%eGB2%2vng%J<n?iRq3W>L~4Dg2nMbz4AhrhTUU3hnUr$;RbLOdrg^AwSZn`e`DDZ
z02M!ME!CqW{psWOKE`GUF0;Asx(8&QceToI4LpK+mPMUsoU&N1^p30%;By>tZ7!S!
zSi!A~C9jYBUA^X#0Jt3-4gR#ZPjOmmfd%Rh^lNyF{jBRxY8!l3maDj$e+%vB$(1j&
z={K&W27!KHsI(A_x3%+Sx;i-*rQglWrh+O3Gp2ACSZ;0YoRxXfy%sn??_s`IR8ny%
zUx0S5esBS>wR8dHVa69)pyz=JuEnxb;r%81J@K-Tt#ig~$_?03l7NGbQ`8*q6p2eU
z!_}<pJruMklM0Jp=O3eBYQAd}(3bf?*Dmvb{h=WY0j#GOxSO=2YJgn>I@}EN$uR|-
zT#iUn+<rtTxeevk-AoN|&6c#W?1krfr>1sR%<$xb&s9>2^Au;iuC);4v_sVo6kDu6
z;5Lex^6dZ)E&$eoouF>+O(nOP5sqW7aCs5@NBQ5j4^$QFBKK;|S-P)cHM23cFa4F-
zqIhDl6FvYNt>vC^?m<jFV3u{OV?*I+(1-b{PLcxmPT-SV<*>;H*Gs1_+ILZ3L9Oza
zYFN2faU0OU``x`f^JMic6O~he?@(V&C+AFPKzxSna9%a=t810hDsNY^jsh}9hK9@Q
zSP!~`@n7AKwL_`FX+o;my2w64{ai7xG|8T#>1!RSIBb2Q*+*}n@7Xp*hUZ^i*E`K4
z8>v`q6Tv^Jm6@C3^|VTUdh1x`SwZ`<-mVSW#=tUYp5upGBA-boRo82GE9Rx&fp17^
zD__axxVZ8(_z<wqX|nvybcCzJNQ^<ru;n=F*+E4!y3}RP?PHIlLexsdILj@jQr1UV
zh3@TyoYUhIz*?mWSL@87@?7U0=yuVH(ra`#RZu}DPpF^d=mafuXUY#~YpM=r_KfwW
z-!aLX4l6-rx3a^^Qn0D~xKkep1A|I5R118#3{lDHF|wmFJ&@phC*PGpT0dDkIR>f5
zLT>p)=_9z6dPc!h)&EjA!FqcYx|(bv_?D^e*h??*XyDbrO)3Y7qZYXB^4)62mIt=U
z80o#_-6l<PnBiH#XT}dSi<7!+_8-&{_fop2dtcfI&{fa}PNrYUokg2e6M)&S5b(*i
zBz-Yuph)|lP*V{1{B+bV+X}5I-H`&oe>PZSTMirmI=gDSJJah@`#{OeCRahKQyo-I
zt-r=~AVs}(r}CJWjCb2M6t`Dtqzys_)g5q>L;x$8hJea*(XtgjZa*ZSqdH0LYG9Lf
z@$M>`p0R}Lo0tcTlI@CB&~0(oGn^8Ky%9aHL@hBenW`HygM6Zbx7VeUY~Z%drQ6`;
zOnUB8=5ObNlxz4q)mg^O)J=T~P^HyVE5RrgmkE?jw*7V-vmbICQ<9*?ekN@MkV1Vg
zZ&<KfR^rHn=DL@HuhN#=j!C-$i)bG$D6TRtsUCc5J>=-1dPYr(t;q%8da!{DQM_?3
zlXqPIzHoMV>^&$OfV3>?selq8iS~}n2{i(?$dxHuiUugIIm0#u#n5f+hO(2TdA2y@
zJa|hzCDlqFaLjOID;Ls`Ehx)!E0nXL1&aTaOUrtt^pMq2=oQ(<zI26URZ&&2k{<5J
zPiv5J$|F*f{j~~gcCPG%?OpK!_;W%NAX2`YsxC>EMU)qkF)9RRRo`o;)nlpV?gPpa
zYP;o{^9oqJ?UW30o|Y>u18le5#{wTquB2a}HfkH@Z7hnH+kvojn4_J&jzfT+S*klu
zSohkWlr&7KX&z*007sbhOoC#Ox~I!38|J*0tOr}VR;BEb;3*ltsmW4$ZOOQ@!Tx#l
zzlrC4S1qabZSsHQtuo_)6#KEv(W+sR6Oyx32-0|lgBidb$45Khq3Kg9yE7SVgFFU4
zf`8juffX?-kIES0tO^_KbK$0R7uN!S4=j=m$XEf)Nj3VG$Q_RUu_Q)MbyoSm$_2Jx
z;M2_bYK00pecM_@&Gv9y3AEgH%H3KLrE|2sTnXYW-+cQK%gfAGt~0JnK%u35+C^YO
z@h^LS`zOyv&065C>%L2=lqw;pN!@sxAGpc9R2C@W>FJJER8u~z8tJ(STvjyF9t7*U
zngWY~H<B&TVns&LI@eAhn+w6~%ciS_D{5v8pm(zC92#{>>OS>Y_0>?IT8?ZrRW<#y
z?4-*}*Uu<T8Q|?IQ@Qr1Uvs3xCq1v|JVk%k9Qj}DJz0ElL&X=_eS4vbMJUx)esUR_
zIjqzMm$)jb763(#V<mMmtjeCsu0?xgPF1DrV4c&74azO>GjONmF<498*?Cneq^z|e
z&I;>v*E*`ec1JpLThKMf1F3qrswCG_^@P{Zhv`2p5xdjg+D<Z8JaNEyV1?_Zs!@4&
zLl^&GI4iRah_5crEmF6Jw-qR%PqEc_+V-Gma|V{CM$!Y@tSw-!-W4jFiqbtxzuN6K
zrT1e&D)>mg8&%5Ffs*2-8~Ma}%5&foN3F6O?xUcZeruPg8rQG7c`5n`>};DViJd-O
zbJEM4jbcyUDSf{C4K<BfY1>VGu=j(CO8VLtYVx#K-MuZnplR|?_Nk6ewvLX?w!5+j
zbV_|Lwl5`EJC#gi9;g;IP%@^KGI65xNVR>`gv@2`V%3b|x_NM6UOti0k!lX8<;KnF
zl0?}UDA{qA$yd~s?H1QS*X8>i4}cyq=U#1jSB=FnQd6*@w)0uNzo43)A<m#fjTEOr
z_6fE<mRI&BjK;aqGfcT0>YQ;c#y#f8_Nux-PrBH%JoSyVll7asn{*d+-Mzjn3vOk-
zFB=Xg<$q6KPUmNyp!01i+n?^X;9<vg%Q3Ay<FdVu3P)1h1|SNX>?f(?%uQI7!S*Q>
zEgR)lBwTkamb;X{=o`>C#YA<wgOL7)Zd%$?y^C`osl8F+P0CrSP|T*^TbFp6W;_AM
zxLUX#WoYfKfe7$S^<4p5IC*(-e0nZ)k$P=SE8Fha#Po(f!gb)y$}=vx?K<^EmM9NH
zlaxiC9BDH&2OL$u$oZ>8<7uRLSol9>7s@0X55;zFmnb{vK4c$KvQD`VXl{$SLghrl
zYkEF4zo?D7w%R2RrwyfkDB_*B6*Ht)slyF!$yQSB%0(cY!Q02H+PG_Z+B&|{pIsj^
z?pbFlP-aEuDc2a+jPg&VwQc`8((P4&jn)^Es?e6q9W^IPW&lFTRC_r7i{(9iA?2BK
zime1V1Oz0nRBEt`Y%I`@E`hIFYq(y>+5z1%3A)qPo)mApw0^U_E|ui^AZ=sY1f*p4
zr%?HSY2{Y6>{;0`W_mF!D+CVM(g7vd#`VTpx3G(2Ez?KU7Bm;Eatu>2_T#o?#SnGY
zD>@NJddh0}CM}~@!Pin7dv*iYEd%myP`}gLIhuh^YQCect(@-aXyoWkWtAV!_+{(K
zoTF@DYhap0boYZN+6gF`K3bc$pUP-lG{ssI>ane}@&%{^Rx6f)%Urcw+vSp!-6eZv
zZRpJqKtshQ_h?m3aQyl)iZ!WYotd&tc7?QO!nllYRBL&r-C^6V+y$SMJS_X|NB}Bq
z66GrQ`@&by+Ts<^TH7auSMGB+W**o^+q-}@C=0M298oreu91-fKByP5ulL-{tPPaQ
zO|GX~t7&egm~Dx2j}x}5(nibwrEjJ`whyJ#Wus-y@=wwq%9_|em)}k505o&IQmvD8
zqAw+0D2<b4+HO0$$p$-uid6Te*uQp3X8#7w)o5|4y-d=j{+x`3k__s;n#qkz+YUpL
zrnCe|pf)?}xenWg<UN;W17*(cptbaD*;v^f=P+H<(niuh=z*4bDc>EZ?T!X7s7!FT
zdTvQ|c&VhX>w){CW0|v#?bo)xP<`NV$@AjzN&mA=q<)kY#W-Kqwmai6ZFAvJu56aP
z82ntkLOQ1G2KCIjl{U-fB+i!&&3GI8RyL-MgU8CZ%DdV|6gGl?+poe^mE)B&9BqI$
z@{gGtEs3rV@+N6bffmfA%rmZA)C^|{2+BK@&+`1wH6_1?ZEg7^>T_8O`nIyz6Si-l
zcK{cl8tayUgk(4%77eAx7dL@@wi?vq;+)vNssy@Ie^IlG)`L}7jCOB_y4mW%|4A^$
zs8q{#$(quryE9~W-<BP*HYl}1ryXVTzm@Yc&y<_WrI|A2Em_;urHTpCUUgSUt0}YG
zuglXUCEyHa9Nn&LN14fXB4e}j0(2A>(j6|+b%FW_^paQD*1Og_cG-s4-Bd0q*sJ;`
zHE)%HFtr`30qq3RnC^-5$`2`&Wt*5C@-@oAnM>rWJWtBi_5_*I*~pP@D^0spC7h{J
zf{K;O-Ljg+y9xpMIot8rztEN>s(cvms_Y95D~rmemsKyFMjf}$<ulPQBu=_OfXz4j
z7p#kI$9j<;{r9o!MnK<#yd!C%SP<%Jd={?27KFwH1#BZfAT(UnOMS(-DcpiIR@_Cd
zYC7O6#QoZ5=GTE%fd`r#Tc9{COQOi)l=v>`gIEvjDL*oFg6k?aHPrO0Oek51w4)&W
zc({6?qcBg?ja;g;?Wk9I$+S$&;n#+Y926eMFV1?bs~!QAFHCX7-AWI3SqI>o^_sv1
zd_kzOGb!65J~v*#H+idi=jvwb-$#csYTK}i6Q;SyPI8pc-n1TXt-~6$#Y0UdbMLq@
zqA%uA>nKiSp9o#>pL{MefvqQGMQ&!TAk*v<@fP^S?4z1|{sfnvRh{{$&k1~`7aAkx
zGuc=1zf@g<Ybs0qwZvZfSt&<DCnD>F_YDUSqq)cEw9psh6n;5&!Iejx2p$amVI;VA
zdL!A$^q7-{ev%E*b7VW;p1K{$K(q=M)?Le*qaTlq<mZWrSp&TH$U@yoBN~~(ctdf)
zk$7{sL@;GpjlA(daNW+q<aJ|Dct0^P%OP&8>|ko3TgP@*h~y-47ChIuA~4cu^LN8j
zIFB$+dtH1LnrTfioDMzCx~#`4)@Q8_#;m$1C5{y>6;U9b_=+f_me3>jRepg%h(0rY
zhMn}6U`Ue+ehQ;#ww=}dL#C@EyeeFu%`>J7k3HwfafXHQ4ShY4V`7~7Io?v|({#np
zfwJ&gaZ=<3ANIlcZR2y_8BK+Cpb*^N*;weGs(mhg=T~EE!{Z`v@v%z3%TFY`D~OU%
znQ0;aR%{hM5VKQN@$}|aSM>2@5kcg#x4rQu_()ZzC^8Pm8;KjiU*c?oH>-y?2Yu>&
z9f}J3j2nZ!rM-irD(&h9c)uM!%>!~}@%ivK;|>2%Js<f%UZYQ$YKB)SdTW~*Ph*#2
z-jdn4gnwp!Cxy78^f>r?q>Z=)>1y0+93OqgHP=Q0@5N8v(qJBvZ#bZ-DZuy(aDBx;
z+WqjL&`GRnWG0l__?^Bs{)MZ?&8rxWya?727OPa;C$yLwq5B$|ZukftN6s4lD6OG8
zow^O1Ntnpj>>|;WI7;`@*c&5t`_ki0v0gd)B9_AhjCyh?za9->qeLZk3%F-YbCqTl
zlV6bEnkm5oIvtj%5%VYWb?GI2ystKU-t0xrczZ@#h(kj|2%UD7D#E)}^YAQmmmn24
zN9XMZja%{M2GQTB;<K0|LXp<u!SI*>DW21}kTzzU;$0)Jz!GtI#ZzPl7Dl!ZQ==`I
zk1^-hCZmVk$JGcoG(L}%>fZ4au`GQDbWn7;#%1hRv4F3_e^ep(aqUVmiu|EzCQ3YW
ziAQ8^pcOkcOltlYnG<6{YI)j_siwX}6GaRD4B1aujoxx9$bneT7!Z)HKOadm&JO+?
zq{QPy%~+3FX*?ImWA_nn#qk1en8h#gB%?;Ywzxm=gKJH`+rAqai~kU2nfEb0!duv`
z@HnDrxYAUUiWl!^dp(D<YlfAP(T4MA7gJDkkli1u%&J9h6}Mml+82K=*5($lhsnmK
zrQ{Lt7ZBp|jW@FG=9l%>$4sPyBR$J|YGM_od}5>|B-_588@j!Pv1<51c!h8$;@4@@
z3yp`c&9Ow|ZM3tt*i?hB85mj-HMDcVL_6#*@mxF4Jrr9lYrd_%fe$|i1_+ZixS^JM
zx&H~8gFQ8UN7IZWeY?U2_(*s@#~3HdTBv={O!X>az0QvG)Sif#vW|ve?M~fWc!_=_
zni+XVmWd04vpf$x`TUje6=4QZHDJzeAzN1Y$@tPgPGb$9;#wMUG0A_DT}lw9wq`T3
zF|>;89vp~oq84Gl@Qtby;Sa<L@s{I|sjiL>q<dZ(2<-=Nb}aWgPKd|8<5z|0S+yz>
z$>-b(;kI}WU*TJ5t-ez)q~OOQwX~PRcK#VT8gIe;$m&eiv#lkA<}}y-ND>(j4JKoE
z9egyp2U}}+j$e=LApD{|vefXqJQ0Dpb>i0Wh7cQMvOXB}<ce^%e0X*$Q$Hl-8wvUR
zAnqSLQ#@ujt6G3V@Uq>9g==Cv(`~FV)kiqNwZi~%3Rx9;Xg8y~E4nuRDcKjCrXlei
zflhe6fLXOMJf7&KllxW#hX_h>dRR~B$ZFz9A}pRI597aeZ*(J#+sM0tcVXD{hI^}c
zM290*ZwLQlK@QMFLhyPlS>zW(!nH<aF+W(xz$N~QZapXWv`dx-UaCQ%rExa;%s!th
zW8Z{c_{RkM8zcI^glzf3tb&*wi;ceE%84DJ!&&Fa9?^fa3y{|4OCTv?^sB&faxk_f
zGJsgA+8gN;-iEmhb%cSWptl1LumOQ`vJ@X5?jw4m8v>n~bEKm2Eq*T&M@j*8xUc_K
zWVOf;qluq(BVw(vdnW{sL|*%oBh9iphF;qzhdS$5M$bX-!Yi2f`rp3w?k3E-tgEIG
zTqZfv(11_Mx@&g&+X^GO1b-UdA9)rQ^+b3SoNL%mw~v`cP3GwpPxXE{gWrVh&c25S
z&`{RFAYlX|mhf^9A$E!M4O{%JiF=`W;pSzl{C9{$;ZpuAMv<_^ggC?Bhz;a4;}XNt
z=%m=1x|JGr_pJb2!_7Lq&X-)?H*vTB6j_H4Xb*<lb0?Ly;38eCU_J98qI&3p@`~@S
zaW}ZY_#B&@^-6aWVYuv_?Q{zMx93k<j<^hQ=5E9lq7yd5_yJDi29p!<v+7gWVX_tZ
zmuZ&xB1jo0yRT3|(kCo1QpTf^gM3@vbZ&MyO8g?~iq|o)%+49`|Fi?K7Gf`OXro?w
z$gohX=^cR`HqXT!KDqjl;w?U-V!CK%k4LhMgV?#w2d4g}Zlon^sbOHGvG_5(mA#2y
z6D0WYim};5tn&ZC9uFrNFA*2TF<gttRo@`B!O)a@BKPVhMWa}C&tu&}cbqT{n=eMW
z+2ka65CWN-c-n}mrbN{t|F`17(VqNmvIln!Dfgsh-C;M8MiCU7M%x%)8>-??^JWpl
zx|sG>cEbc@YJICjhV`gCLDV5;XMJm=@W1lS%-Zi;f)zq8{as-$f7{d|JUejF-&^>L
z9UyKX^2~nhh`K*|AXayuV=3rOe|h#;=xS_tJTD#wzlMhqo5G`XCUyaNP;oePom{Qo
zPDc6Jp;<x`Q>yWsjzh0U^wC>X0<u!z#5UAp&C+nH?nYo2vfflx*crI3ZWUo!9_!AJ
zG`0-%;uCcr_2<ZMvFT+`0|$!c3L*ZeC<*PS(=fTYKbsZW;7u~u&w6O=M{bJjMur4d
zhj!Q+!R4+c+-tI>?hQABOv2=3O|6W#Vim;xa75lO)CHTXJrl423i5xrF3R!EB)QRJ
zqy=j(E)q!?G?!>Qb0eeK!MU;OF*CGFFXgudensyQ-MPkkmH0LG<$uCU*=7+jc5<7*
z6mpYo0p~q_UeHu_Ham_x!oM?mV@Bm+feS{zFkdI<o~loi<J3s(u3}?FgBZvB58e}>
zRPkAday0&vtr3-lC+p7}{|yja6>=lm%wMW3!bEYMe|@;SzAwMU-xBcoF6;lm*>D(W
zfGWj1A;xqm^bGI9qx|T|SFI;t3m;N%z^eflBNK@ALXqbKpXq%-{1eJU5Hd~N&&{UN
zoudqQ#p-Nj-seDJFgt5}6!msA4kNAXNHP>iXMVU}5v`)*jEQ77<0$fs*@PX6ELFwD
zjBV}1V?(R`shZQkB@T_$Bn2@NZX?MGCsxig)*xO8^Z9S8-4!qW+6aNj^S%qw#%Gve
zL_34he>wJKZ^`F;7u8+vePEva6)iV^ET1GiM-qfKMx!o1RHok*B6vaP$H2%C-P7RA
z&`rKR;tskhH)y8_Q@N-qK9+Qd4-YT~xiz74dOTbO`V$`+BO<<pIz`{cR#_-;6|ZB9
zidkaBKQ6pJ^bc^tIw3qXIu$vTrN*<vlf)<dEN^FS5jHw}3|}Tb&hAHch)#yJ6%9-q
zjS*u5(--|?|4QW}ZmXIL^avah84u)N%}FY5ik(nh#eTCMUus<?Hf7ryXJ{UCXG1l_
zm0A5w2^v-msO||<0;w?%VYaUyZYLi{Os4js4S~18%j7q9AX*D=Lq6aEfD(@Az9aIA
zh2$%J(-?*EjV#1>SM(1TG7jdFCQjX&WC*XY++^YM@ap^p{Y&ctk_as&I-?T32rz6@
zp)|BSOA;C_W}2pg;h4wwZACX%$n;f}V*tVtp@X4BKuwHorB(sV(yfa*t0%^CX>&rK
z!ip?uw5jVfHiG>bzD)G-9yZfWd}5+L$FRqFNZW%f);~f-qyX9E-)E>vyokd3%`tPD
zmfhifz735I13Q^F`Q~`G>#FLj`!A9$Ds~d%$s43fbckcsJ4lE4Ong9JHf_Q-hi~{H
za(Z|W=`sEfTN+DcjgD%v{xYVEUhy&f*_lp0weyYth}m)58^&c#_Rkj;yjR$%80haS
zj`j`K<}xpItML!I>`((P(eyHMuEH$;9J*ckw`oprwf3;Fc@`v|C%c<a;}O$iV{>+%
zF)v~>3i^kRs%)9zBW56Qbc{nGw!v@5yg)zriWm*u3u|Ovgw_bxkBsx(CNp`S9V8q{
zb%{eOrVD%gRfQ|;D8o8DNmm``ImEXCdk_flClGsZk1&?*ZxakJ@j*f*ciObB;;^5=
z*EC+P{3FIBg@q1OZ?U6fl(~&xpY^A?t1sxE85|-`HAEv)PUg9UEaOwN9_b6+Jp(s=
z&mxWWLkt%7KCupr;xGI{Fp9=S=7pMs`*VG@vqKf)YZNkGGv(3sv0QT%l&<U>NyoOL
zn5Urz3pEB`D)i9g@HL_$cu@b`&?d$RRgdM`akeJ$jc;K#njhm&f)gwBft5%>+5fUy
zqXBjc-oaci^xBfh4aC|+FJZqu+N_{{Md%l*h>Wisl2tEb0)IaA)8E|iGsZukK^Ez%
zAiGTI{(Pg=Q!~`7qOGaA=AyAm#0T{g^J0`na%hTBU8m*U;u8NAu}UUxaB)?21#A!U
zqj!?CAbQs4z=i_XM3JLH5_4JUOfr_V@{b6q(BEVPU8?z@@ewcOO8>!#l2Au1n9h|O
zf+J^yM7#rL;!a^X%m(oYHMjDPSBt$i%7qFJ^M>p_#fiac*kYowG27f%^9Py;$7!1|
z8O952Q}LD}?B5)lw6RN=#@>n<y9Z-Zak8PStzOI$w~x3>N&{oE#uEjaX0{8uLfLC?
zd$a-5S(oan^qaiR1u6H-@RIA}KgYe{NBL^=nOvHtzq${16Im6yO8|VPa6gtP&*uN8
zUlT9Hr=b&QXJR5Lh4tj@@`q@fn17=U|6iypFhym<)``u7jIs&R5sUa+^CS5nvBG#h
zbTHflg>_c0w^3uNtuq@q_Mx|WxE{8c%a8fQYkK;bUTf!s>jc+^AM4J<sOtVieZG!s
zWuTt{Lmr~1Sel7O%>1jEZ7#<zYL(!{4UXU<&oumr$Eqo%YK7b5Z9=sPy+dvIQ%Gfh
zXbog%J{}yvH8V~`_8?1Cfa6^7XY9n8jpul$a{J-cdJm@$&PSJpmST?#U)<k4>$vl-
zOQyDAS0s)wi^t$i;Uf4nY7Rk6Jn;^%>QT8i`PX2x$j0bkzO4bk7Kg?Pia;&&7WzEc
zgAlPI{t5b?8YBG8?Qu^~)bpLgS+#}xh}~nw*ksBohJollaT2vVyaQhum|~ikzClw&
z-u5WS#fonnXRYMMS|{+B0rM>}irg3Sq5e%ESv&>b;6-#fcG;^azkok6?#14rLrDcU
zwLuMhO<3ycU8Om1;cS6Y-A=t%KSB3ge2Om(4AA8Wwa7GVt6GoD<fM^QxBy$o9YIe9
zZ}DphucuUfiJLAg^flnFc)ya*g-X1(@HKdWlmzqec7ca{6LxWUeQ+qU6MqV}#Ak6z
zbv^yR(7_4|(gQPjPWm5Vc}Rw5AXg20>v``t^OsRpXEpTWU7iSj-2<X#{<sI!Peva(
zuRCu08yV{QQ?(EHf3UN@CL*kR9Z&^k`WJh8`PKw2Tq!$1mnYl}kn|`1Gu|@X9nCS>
z#D6iRN{*Z}^hdwb-&Q3f8_8ucpN-i!nZG1dM_1!F^)XwFa2p@vOZ7it<ydtjh16<t
z#TV{3&NTf-yar#7{+C}UPKa@r<0VDvUWNeNEU^di%Sj@7p7!4tN$wdz2M!6nJPL6=
zx=2lm`QcyEPb3^;N8}+hx1W=6r}Zd2Mfj6nh}*>8K_uKcB1J2>qj-j{6QkC7;TRLg
zmKR@SDe+%J1I>5tA{--D!%?4^SrzWcPex;QWa<lN6|y~>j8lmRMTz>B`jQyUb0a_1
zI7~hPlg4H|u2&Ai?&?nYA~9n;st>4|VP@<V@hnOOK+SG!Ld>H+UOXy%S05o~hZ_af
zh1xN1wCw}+z0dJS*xBG~<QRnmbC5aq25?!@ChCFxLh84Spr>wpK2RS3nU1R&pq;w~
z-N)0}_l~jn256r8)4(3=c<*wyC-OjE%X^fvF#)?n+g;rOX-OEhuYJuZH#@)lgSHrc
z?EYeZ-C(Zr2fbH|GXj&N+2~wo0lbqmoMyNBxyM-c19}4u0~SF&*mKN7$HuhD)Fhd-
zSOQ<z_<*{|{FNwGUx3~fk0@>EoCz*SI<Cr>&$l;oJ=J~<wo`PZhFN}j*Hu9hdP837
zFZDvjR##p5O3fs;X>w2Gx%(Wvn;y-q^nW5ko_M&QeS_>M(?h9;)XEdoAHFwC+ifXa
zE;U#;T&-)+QhiBN2kwp2?hQ;;_+RFmCrEEDUx4a>8S;?4ru&YgF4DxAtO>Y2s$EQ)
z-%zv&EAf!*4t$bLs+TA$_4(eRnhZM+ZKWNm*W12pvK6bXt-MDwXQMMzf5;j;y1@^b
z9A=Oo(I!<sQ8Uaxj#?lQ+Rn~UZBmb5s;bEvtaF_*Ui}X`cH2Dat+q4ukXr=32mfVi
zvU>Gfhrk^4&qN0MwDxl3W|>LRQTGq<#?=ToWuH%GYnM^o=u+=A__n5vY>D=f_ZEdi
z=Q1v&bWdyJ9ZnU&x1<g@L*2(ifCJ(k%of!$W`sKhT%g^doml@mz1IJ?WjgcJdskY`
zyL_9`H%ooWx!iptW=MGg{Pdnvr+BV=#)8RAJ$g=A&luHmP(2g=1-D~P11pgO5>akM
zinQ^8G226ME9#u3J^NN2QD)PV0&UrWj`QjS8iIZ^JKY$yjx2&^>ccEeg&Z|3G8Xat
zELegrL=JoFyA{O-x@PQ64B~5fZ^~W%hVX3LI4$BzQ!D{)I#-l^%xt9i#Ga{lPoBd4
z;XJcVUi=T-THj2U$Bv7vP=AN&*!p|+d+I>dq1CGX7EV3cQ=*Lrwjxrds%)t*13FbM
z!rlD$+1%P2$|k1IqCPVtpjz^Kx%H{qlqrMqRtFu_>Eh||3)>@LveT_PgFN&M^?ZZ2
zIGmolrM<n=lNe1#LB4H?-wX9C8KCIrI%BU3KJ#n@`#Wmb&y{}FxhaS)0iwVN`m|~%
z)DD?08|oVc)w3@G7f=X$&83uWLslqB?K{n2_rLPv3KydQ<Js5l=H>g`qrLxVYJh<4
z*tYAAo$#sBb8siy61pYT2C0EOSH5?Z(VvwcRj0s*-d4)^vd%jn+43P-Y%=N=zXEKG
zw3GBj{&Y;V({v|~g&LpqNwZZB`FfR)wLeoo_Uw&LDOiToqLqpxC2N=ip1Gc@c||do
zPjk4vFHX_bQ|c&Ux_FY<!Np5lr=8=dO!&Q~vt7rnuyvOI1ipKw>s!P!QFG`H?oO_$
zF+<N=w7<7lb2#x5(6?+D67+1+#Jmc?1xtTbcjmAR_C8C#XFaNZu8J_FRBPJ;U>Z;f
zKLqc4zeq@1@>VG`oa)CUE2aTNXsXJ(JzJGXZ33Dp^=xp>8#r5g2)!h|Ut9*|I=|aa
zGl$`hg@g@PH}R{aGaN|9YL`s24e8D1!z0T#3x(Uxq+VunJU&LLTjF}-tf}e=%+dXX
z{}dtTRPC;~M6cfU1=&^j$~p*m<R1@6^BZS;E8D1kMQ=>m0w+PU)02^2+MM)fs$_MV
zb%J6oL|Shot;l!;M3kj*7nM3D#}O)Dks0f>0cFe=g-(}giMJoJ{Gv1IHq<Epn#{9m
zgO;@X@MQWDkhSR96dFl)wwGN%mup(8R@vYCyK7dLRbf0E<EfQNXo_3@+@0<+tZ{l8
zAlcL$_B?Y)-PVyM-J>4k6TNq>8yO`y*1ko(GUMs$e^e=6ouBix1m+~HON^suc*B^t
zV5nxLW_4^XRtfdX(=|Co(nXtYN9AS8+Sc?~eHzKgu)XdkqRVkYv(z!vxl^5venjY^
zhJu6XugO}`0i;qsqzM@VnWsQ|-D3}z>VuZ6dwMpbCA#N~Suv6MkGvI|U2Y}oF~gXn
z;16$#X9Sg|{Rkt<%a*C`>%habyQ=fCZ|0)xg}RXWPrDdsq<xNbFCXMy$m$S9ak+P+
zyGWTC%JzTvm{h+t*}y1YI>5MUTW%<%@D!^8GPCau&5E8{B+vp~U8Eb-F;EZL5ag8w
zmaXhG_cGhBqRWv^>c6z@<!!0Pv{wT{ACu+=I9jH4g56wgkrG*@rlGEG>QZ07q*|3j
z6{9@ekd?MVW}~}yBUM)1%JbeoeIeihc**fJW3_IF>pT+xd$?BG<Z-2HiSHQuF9WP-
z4s>E(6hC%+q|d7ks0w0p0@_#;yaR3hRDHEe>FQuDQ0?)%hdG1Lql__@=fomOk>;tQ
zuPohDr1h&WFawcu-aXv5z_yIh)Wou8>_)Vc0)szD{xD2cf6F-REY|#D8q@8fvpkz+
zx5|$KcQs8Zi8GuO@88ShQLPXoQxoPZta8?$1q09r#c=P8lEacbjiugNx`|f+FKClw
zW8p8J3GAWtR{)Oet=Bqp6|-A2z&kMJ@javdfiz&|Wi~8tjQ-X%*_sRt!It~Zc&90x
zDwF57dnkQMS3ha7c7`|T13il+!+>mfxW8O|Byk9H2)-iUrK0TfWr?=;U~Ez+bGhiK
zroW_F%*DJ&eT4oCsi!`wC}O?pnQUVG2XC%tg6e!Gq)B(LR_s<Uq{e&CIfj80Jx-Cy
zG?vor|Fr8Ae<t1aTHTY_UG~AA8%z<{ReQ<&97enY{DXm3u-EZKJzF7nZ<E({ozOPS
z^{PI3jsa7*R7XF1qjaJs)wUnr=-mKgv6=6`VW;N-v}f&k+j(0D<cPARD~*|quH|Qg
ztDT*wZ|)7g<uMn{GvKOw85pnHnSL(SsxiX~?R@rwW(qW1vyA%3-47g6k)?i}2FX`~
zw>62rSwYrEE4!<f(>>rF@IFeeI_m7sY(N*e_5pR+3U80pTmBVRmsSG(t9hYW3xCqM
zWH+b@)JPQw*Mt86a|0yO5`o!r_HXx3hAHzRf7)7=T(KjJ+1moxLBCPEd@a4R)U|`B
ziq@4DNoybzeG}Qk>HkRM{P#5*;s--AV<R}MtZymz)b)&$Oo`3WY^7_Uh-B<>#0M^V
z75QQ5ZJ1Tu%NPJZh|SMddNs-g0PIB#(>(FcdSwipDY<5e#QsvO+Sarh>BFuDZz!|i
zv64oPX%-jrueN<zp5;J!XWs>T0{njsor!-;=NiWwEfFHyWRh9VnR(ATX9yx8R8(4#
zsD@I@<=Pr82@)x-H55Tmvdbj1oSCz|XPHC<MOs8Pky@&vi%Zqcbt{@yMJfpH{0Z;p
zd7sbcc|OnY_xnoLb4aF~T)WBg0sXV^o8@(>SRN}{0}9qf_!jM&V}@2u)*IR^bLiE|
zm$ag-=8p4LtwmX2_`<(}rSN?IO`#0=Ku|!Y^-prYDHdBPd&15H6dv~qb_5ye{gPlc
zkWUDo_>eqeOoJJQKloJoy!cdTRZ>j)k`=z!h;R9@Fq`rWZG&kwwe%K8w>FU;4*PL0
z#CMp__r-Jn@O8FCE<`?t)$=d-VF68VYk*DFQ^^In#weJ=%of|lMEop~MI9BF3-=rw
z4X3d`{cE7#wg9E7Sm!%nH(;>k;wHKUV6G+jc2h>}JULft5rZnX?U9KcLa_RrvdRCI
zZd$fT6NKuL%lIg14SO?|@Sk!uVV$-%_6hk>)MhX<d@A~f!V-(suc=E!4wA*6pd7SL
zNaK^$q4FEn^-6Id`F9i6?050;=G*2$N}(Jf9e}OQ1Un{_n6C@%Y_aW0?E`WI{VNqC
zl?T9gMV5S5Kemlr=yMBY{?D{T|15Y5;mlV+EA6$HN<pN?1mXM8BN6AZ-Hz?bVQRj8
zCzk+rivgDlbA@;mNGyIu9D|FLQ{0^Zmoun3v1qJ6NgeOE(-Wj91r>6!YyCguGb{lW
z0sKmPKt)=t)OkKzo^9xhc9YA*bwWI-7Vol!!dzw}afDgsn+IEbh4^6QRNYAVQ)MjI
z2Pwhqpjcc;UNSqyMeusycfE?XLXTyp;Vt54_Ege(pMhB{q@sSYgHpA}f){z6noph<
zXgo4brzI$Ng_nvW=3lC5&sL4GrfadF!P{Yp#5ODcp{nto#(_cuJ`7xnnHRfVIBZL?
zoZvAvyyl^E1_R`5BoST}>xdpOluFWGV?Kozeg_>)hZ9|vrJiN@lRdQOsE}d^(0i3m
z{slPazsU?_YfL}Z4378<<jVW;`3Jrb+xS3h_^<8DRxZL;`5LX~Ck2?nbzi>cUsWCs
z;sWf)5G7#8z|%lFGYKMwg%Tw%2RV4R|0+H}Dg_x#5}zcyY;$Ft&%@%dOrjIKU*BmR
zrA(v#F6zb1#O%INa3nT}KSQ;mt>!RcA0t$~Z~t1X7jE#2)gtw!F`tZLiojDR5pc79
zg5BBWP`{%K@Y_-XtXJ##g~D2DFP|g5qm<Ik+D^3{i?)VQy+s`!CBDsWpwIOL{lXbP
zR<}PyH^YE{b<76;HsdX3wN$E1fwXtBXQc3*Cyfg1IBF9!j(kowfjE9spzhPWms~ky
zf%k&P7)u(O)MQv=o6SB~Zq?W=tg%M?6saW5Z0e3agyzl!=V1pNqm5xm^bfp=?yX-i
zw6Xp^$zkb67Mn*^*!S9surBL2^ipIFk!H#yw-W94wY7Qj89Ftl*7}Y1$g&canYz5r
z%2%S-zK^goE~Q8S@^wN9=*|lb-|0v3r-e1PzttOjE>h|pM{i-0K`dWMHDH&^cSb*u
z>+v<jkL*&-thJ~;sAkwH_l-`5X&P>Cv_tod%FT?PYLPbzvBU-ME%ger&$b+CBVw^i
zWsQ5Arw!Xo<%w6kEA~$1?R>4F1V+Kx-Yn1|UsW1CU*X*{Ql+E%s}scEosqV8xdHG)
zh7^V#Y^1kR96pcGS!ba$P07SW_z#`$8Rq7zI)F(!>sw82v42bU;8rQqxq3sne@;ZD
z%rRN^cd7H-5haM6K*r<Q{PDdBbQbtPfyO*C8b2XDlQxq(KrwyK_7*5MKf+stOkus$
z#Bb8hF{0+fr_+agH>7fNpZK8K6VzvhtuUFaWMjP7nHe4{@d`w92Oal>z=;Cxw*E!8
z7oSItdg5I9U_TKIEQE&kGIa2?H&pC4{Yr!g1~`k&wLJoZt@*Z(eIdjPYd%wsI+gFy
z&d^Bv6mmFr+GUbLWKq*m734r>8F2x*gbr8B%6_pox;^x4Ee*Rmq%T#){32!>S2^w=
zjNuJJ=bdA7+8T*&F+a9ZdF+<NZ^f^<X!1RunS_z&eS2VkF+#y;oh#kg<z0nN5FQg}
zT?f!I`7<(I7)acuHVP&3bK(tcmeS21_YlhDY6O|=o2KlHwu*CMmg{!3i`x?6eZ2yN
z;SP);<{B0|*8u}H7zV|j)ThE5>JkSB^Tn3HyN$ukU@e~R{>f5<f6Q)X-ChRdg^i0@
z$n)Mq@(`)Fd0LOd+&TH8t%D0j8?Zdb6#j0^Ay^+efdeY$$i^jB;2Vhbpr3dg*+~|{
zPN@!mpqmH>D`D&><&R+<dlodxCDms@DSMkdCdEm?ct_M>@gvs|*DzC)eyZn&I^3q0
zM^G!_aASqg#CqftV1)a7X$suosPOcoM$-@Z?bKS>Wq9s8rdx)_b5|{)L?}(#x;+W_
zSgOGM!Ztt&Cl{chLLzz4TDa$_d6(KLq*W(?Zl))eY#mb7Dt=B+HfEwx#Y>8-rRhSd
zd;lLxmQoMUoxVAd8L(3ugzVIkgoosI{s1<VEU*3$&5<W5DZ)}@tLX6Qv3zW^;d6_{
z6id9otzxR&8y{lI034qM5~zqWy&9=otqgL?-nmx2V<Wze8?F^f1=i8b3dbm-!M9Pk
z%)M<Zlvmm=!70`PYCoC?MCAwV05X#|;;}H?lw={{H)sKsL0$_aWR?lNl(G7MqwDOG
zqPm^q$h-JSY^D!WTVdbIGCo&)NsckRiJv4|oVYj&#z}bt3c&~}r}B|yEM_-vk{_du
z5W@#ClURRX(>?<4tCOka5&-d7E3Lzhy4DKgE8Y|bDpP&i=^sszl?!+kt&<L6eAQ86
zDrNJ<$^Szd0y9%8JB*GoS5wo0k6daQL+vGl+2*PouFaQEe<jw_{h19tvXNcLXx|Y_
zmuIx$8Cek7$6h0jD%!>KVl<D@6UE;CZWs=4!uVbnVYeBYlDxlxaB-5p%zM{zlDt!X
z8A~KjP;OH&m<+a2E9K_i?Xs!lXZ$JFE3%x$$#85@bPk7GUyA9rJp6(WHUExmqZ8G8
z;!mz0lrH2;K3Tk2(e8Q(qS^R}xw5lz8u5TWA<UQ3TxR?V-l2QcFP7|bZKxPj@%!#m
z!a?xN*o5RDrM_{*c(k9s(iI|7+IX~?Y?rHqOw&qB3Lc8K>R&qUi~Brtuqd`K@t0|d
zkykc=X72#e?&O62*y6zCG*WDk-*&CH_GjlRa_9(KEE$Q<iqx5>na3jo#Hrwx`y71W
dzQ(;zxZ!(p3Ub}jC0{7&Byjq$_rBc6{eOlD|9t=e

literal 0
HcmV?d00001

diff --git a/wasted/disco_test/3.npy b/wasted/disco_test/3.npy
new file mode 100644
index 0000000000000000000000000000000000000000..7d00dfb6116356e50e2bf758ba2b63438fd32ed5
GIT binary patch
literal 35968
zcmbSSWqTC&*Tyv=n`E<D8JFEzOYj20-QC@aLvVKs5CZY6jLYuKN)p`Ny)Dop1&X^B
zr^u84S9s>#{9eq=HRn2V%gO8BuUpSS<;%@2x5zg3*I8p`+L|D?rsL|`T!?L4`pnrg
zM@<`%K6C7^|2uC#YRar%%g$#_7&ZOZvU{qI3;D%`)NRnPw%vs+LH-|Cjen@6;opdQ
z*ltG#GSQJ7)$$GDA)W#%fk&}_ivlU@C^vDj+$yb{_=~fGHJK0le4$f;jG#~M7U*a8
zMteG1U=>`IORfrwg3BUTv}REeMeWbD4``jxEHc@5N_}R%O4*tGV0&AZW^in>^-$`C
zmy*v#GIGg20BL~slm@5~;F_9C*@JO#%?K(=4v2recAev{tx_P{B}Sey^PJ_W8;-fI
z_VjpkAa#QJJM=hUa>d$U_oKjcwGw$x9E9vv2f^8{x%kE4sVHy#SN((N5WH%etw^<(
zgxaDcT0iwNHWomsq0U+PfoMhN*Q6b~{R#Kou}U4~DO0V^T-$N$3;93F?Sg`xypN-P
z#~jy8tx}i@)~DyO&x4iJ3ifyz(FARqt9c-CtKPNN)|dPweFXc62=d(5np`KmbHu1c
zfkg^K&es0rPy7BP-iE>iAXjk}aFx|XQKN7hU4fK3x3UN9CPfA10POA)8llAt0y`IT
z=D|W|Z7B63qN(+KlWP}Hc+up@2Wc{K!?{Ie+)JVn_ujx@B<Nl$b&T?%FV2I`*H)|b
zU|^Iy*iMH@;1a*yzAY5*Zls$bzd{yCCrAU<G_^%!A9~IWktfxRVngH={9UU{w%0y8
z1_=GsMDe41*s(b%5+%YlswF!wxQNOPG*ZV#Kl|2(-rGL~j(O%;l?YAt3tZqy7hZQk
z=$gzbalvxIf8ipyx9gm;l`rp_LKeFI;OAPdDu3A`_9C)Qh@+kru5%1R?*-iS1;;__
z{=k~zMdBRov%8vgNw}hPAkvR~qOnd)8=x}wTA@N>mn+^`O|6G{<Szk_0tIK<`k{YX
z^#RO%A~M%mUapCsql)ljVSy+gS{WG6=i29THRQ&Dk6}XX70s~k4E{zfbFNW`syP%M
z=t7OB-YVUqRn#ZJ>`0Nl8B+`Pa)(_HWXkncc?q4A1CE>8^hl8Giv1B<LFNUJ;NPKk
z2x})JA>yNU$3D^Vn|21N;Myk@yFOe0Mz0swbPaX)#eLTL;V1F|N(?PvPiqEcl8i)O
zsFmbat{lROT!_0Lw<emY2~s<iR&Uxnhr3dK=QdkMsZhvMb;xS*MoiO4&~Y5$i3!&2
zvfYviFQS^Sz7rS~Ie|A-df6w0fS^s)xo^8>nHr)SQ)-60IrdoF5KXkS-~+NRTsw6v
zHJUu>;>k(cOq5l5LE-LDA6%$oh>^3GvmXVo2I;^lqBX(7cDYy38Jg^xY5&LkAbgs9
z!&_ZnT~lq%L&?~`wq4P-?zD(S{)iRl7ok6*O>9P@ml=&yB9ImfUyjtr_iAHOT0k`-
zI_-gKBPJpXQaP!$QZY1u&nS!9Asei|aXY=~WqC7Q+b?c)=ebYg(|jx3u_48=R2fek
zwKtKxuHI^<tFfjA^(vujNDT?;T^F=vT1O@9>MA#kW@$}=GvK-!<@hDF1Zj?pfJIN8
zU?`e~*0Rs3vNNO?E(>n`MR+VL;$3Qw7kh^rv#+oY%IK1d!4$Eg9Oq0_mf10~w-RS7
zA?mrZ-R%Q*b3{FXtRhxwB@r5=9L7{F+$`8pJqXNJ$I-Q2Nzg~-6?N8m(^exipYJIA
z3<$OwniOc8|J3`@os3LZH8n1B!qGy@p}r8asRgd_&>2@8+*&Tdkd%D&oSNv!<PAi1
zd!ld<ptZV@HzKbTQ@g~=$RX}2wNG6c_{S2CwzIc%)T7ErmXjWSor>5eIx9H}!_%k$
zxl?Tx0i!_4S8QD<(>@3d2|sM_nckr~p-f*}WDvJXtsVIY&jJ{6a^!OP9^{|=JLkvZ
z4KC0ri=)*8!Ixo8qeG9dqgaHmW`?!e!Ae>q@<*!9w1W2$UAR%sBEG6`nAY8Xuq@}E
zXd?|(iz*b)=dJcBZUg>Nx*x`o*XEwU3+--bZgFY&*AS+@F6|Nd1me~)dD+DEL?6K`
zC)8eKWeT>g5UQ)wsGiaiqE8^v-oSd3ek5Z`T;#eVnd@FSP|9+i)DDD(lUF0w{8tfu
zk;#$A?LioOCN;-8C~%j$Zc}-kHO&Fpx|B9?)lnYcPU>*b9ULF{PCX&d7w@odwgJ{x
z#J_yQ+B;#A+Rhx(Xlsnn$I%YUBL1-7Q~M|ZbRGFYK9BAvR-|(1R_q92Qme~^kS+d4
z3`{w<?MAqq?Yk<<rSc~Zwk;#d+2*=3BURm50zmF{Rh5@GZc&wlLCRnD6WFLgvTmQ;
z)jCs4r{1aCtf{h7P_%y3L;GRZx=;@>o7(~Bic^9Q0>`wF{iFS-)(=>(&K3Vu>+rSY
z0q6!|CjQou?u_$Atc-J7xIQ`%4~Lt^c&O>ZTmFf<QrcVh1wr7yTOs>!`bua}%2M}B
z`%HJ$S}Uo}fqm$E%HtoGYIVNJ2ORmq##B0{2G@knIfl8*hu#J^axckt!~=VE`+8?N
z>qM%CdM%KZSY&(YUMbAC?vpm5pUC&h*~n~o9kn5}$2BIJ=umR+h630(-7}$}WKqex
z602|wpAotX6$L(tQ!t9TOFoM{<Y`&7Lyk0JAkoA6hb>mzLYyMEW7)!0=h4V(qKWHX
z#2jiVe<p^=yY0!P1ob9_S_f0@0~D=yZ$P)H>j@sxxv$YDiO=?5f;Hf=)Mh9N3|co5
z735CbpztPjpY)TM#GR6tMIKo@tN*zE6B@~pKyUkI2@Q^h8&RFLk2Xu;#n3vjnFbNv
z;c)OkDAo46<JZtEdY-+y?O<pJSygFRyk4NS`*|m@J9WAXm*M7M#95In7ibc`#;@dm
zhA-Q8hl*_7)Z@ZZc#gLjxzo7>UFk3odSXc+mO7`rbnJ>|hu&g$Qri?ylY)-Hjs$Id
zaI|s?j&ZIjZK7>+w!+`IC&U1%NB${ZsU7F+2+a@=N`0yRilcZXDG@!9e~@dgnaWJk
z;EaXxgEiG}S}$fJ`U%kyzspynFGHz;Y2;V+y8Q~a4X>*%qTUH}wR6_9@{E+Nq4xG#
z;C8AYWER121?eocPEDj#(4aOZ|HB4pQ7d6@V%rxESS#3`Mr|ot@ebA#fXt_<F(F&z
z7I}oe5UNws&NZ!MvzSFO#zjgY-y^y#&@)omE`$~%-=*VHp8W)ssf>Uo%RPcl|6=zf
z!euXqR<u^m>7|R+zN`$ZtP4<ep=;4Aj`_~o_Fr6f0<xa5eJJ`RRFyAvzLi%5yBjx%
zpVdQzQSqQVc{$KbtH`AXKRX^dYxtC;mW~(TI@h1p@8mI0IbmXM74~XmKK6S-FRDA(
zUs<jN*r~zkR2H%_MGTywR9h`|hP|Rhr;O8TM8_&AwzN=^1~EQ)UF079Thgp>U8GGo
z9hpY03aoO~AvAco<GmK2(u~@y{ZPxHA3{zg-tiBbX3q$hQ)A;Vkl#Uz)lC!!%BxP~
zIn`K;+7}kjwu^)-c#fK_Pj#2%%#L19eQ4E1rV~B1sbaB}4~$W-IMz|?sRk4sohp0S
z^n}~ixzSd^fzcnHyBbTZSDm4zHj`aXq`5YuhnVZ3=foLjO>rQsL;kZI#ny%de6>~?
zkD5Bw{O%k_WoyppJ*Q6^V}IzFEMx)GgBj5ZRMlv26`?OiVxzZ$Mzs}!Nsoev%}4LE
z_LAqJtK|o-R{U$Oh8iM<`@Cd|>$TO5JaQew*N9o*7%JI$6u7~UQns3P%6n%+R~u_P
z@TFo4KP6I9KKPRA7Eu$_H^CIyg^o}UIlR{0ls<SeROG@GCb(X_g~kVBBa7fujsx00
z`%Lu$(cXFA`GD9j!zsJM%^iIsBb<9f3B)_7e&{dv2(7=fP`-i;jfBw-wV#q3u^rAh
zwJg#ugLANToc2vU!b7e(<XYsMcN%q)td-h|jD&k}Gw^e^I5%w_WPhv0M$B4E>>W6h
zeS-OwROW{>-Px-M2#$|TR;Sy0ssKIKwjdN%cRDJ=IZ&g}Y^@iy*ftFvXRF{ifDaB<
zptq_+iP_{pzJH!R#At^jr0u?f6ApW4d8DnK)FQMcXb3>|1Of>iLehv5tvfsb=t4D!
zxlcZFD~?2<RJap9@A#WsD~Z%>yhF5;vy!~i<)eO8+uBj+D1I!I=(`>LQ+wb#9`*<)
zJ(b*KNlUt1cmj4ooa~tFS{poO`rSGSyQjtpH#91^M;#F^5g4kM^+~{$vW7TI+_Sfo
z+XpHtHv0msyX133ZI(zC!WCkbKv8*eZ0KKGXO#%=Bd=k9xvQvA@|Y|JOV|olqrC?H
z(ly`x4^vm(1*J)IB2Xw~?~iuY2<nMv9Gpp<v7V#a;up2$(mL(CwnLqvjw)#-o~On+
zDub+jq@1IqD}5+G{JZ^Kb`Ncgc9I|+V;ziemhuG72mW?kbhMV*B30DwBqD7o&J1-7
zJ`t_818N8LtTw^*wRi@9*FHu&A&d=mcUqM0L0e>^T<WU8ejsm!4Fqg2@2una6wM0G
z)2;>|1>+)>Tz|Sd5wDT;)Jbbs>WKyfeg^)LJJD}~_1u%OrLJK>l}L)awtPtFoyr1H
z+Z#JzTcf=Tc5)W*V>~~=9--^%=HfI%JN2{eW~2%AC44YiJGfUlLk?j+27khoP^I{{
za4WWDq$$yg>|(2lPbT$fJGEA@AF|H5(3)bur}c5<QkZ7u8mMk{cF3oFB5-I<$Qzl3
zJ;Bc?h>+`gL@X!X66rvYI14l_!KwD7FVxoYLD^=#FZ}LyQl*g**89p9pv2(~azR8h
zW9^0O;&No9d7}D0M34#gA=+!KfqG9F;&~u_VKCylxg<Kv?d7usZ?vXz+8%Fx44zfe
zz`Bu!;05)wbH3J`lwG_VCSk29Q^R&CYj0>qII+%eR8+eVTrQaH6T&Zqx!xn_e9t^r
zH@P0nYDTKJTu~YutmQ6ESw}cM7x{+j7j!b^L>D`qa-&G6ynd1IuASO^XfD{Dd0^TQ
zAng4Djcpm$fl}=V<V=lx!rst)Xqf9F^@I9F%8R6AzD}u4R>g<g&y!8%U3?Ghr8-1D
zhkYk&Dz_}J13klwnM(E^!e7=n*T&#6>>xQ^-YdQ4f44pgeF&V7ZvzyBje-6^Lor^P
zobXLeuiiIhPXt1*AWs9CL`{W@CWV(0E2PR~ZTog)tE(Vb&)HdgX7A$YDBaZ0knhTT
z@Q`{i=nb7!2FmwsMfOl|g10~R2;QPSgjbQrayjjeFGFde9_HJiiGfSnIr}8CCh-F~
z$iH>!<ke13@B&p2MaWx1wkJ;=EuWSLl801_;{nl%AET`;>pLM&8)81(H~hPCwz!vS
z!QO;V$ho1j38U1c&=i!ELPBG1Hu)3TosemZVc$_5or-Oi^=#lkcoY6e^^rlLzm&tj
zw9SP{NFd*{MZ`3@5;w&lkc57{)L8vMo>4DHeh}rIqm(AvX{;C~hz52dWg(sm7rJkd
zZ)~fP)53V{sLg8&k^MuD9PddVvk1AYta4mUJ`1L(71TxHBL2Kx@5UmBDHnYvn89bb
z2U$N_c2hO**G7FZ66v8b`8|V+Lk!20Uim-!=s+Fbi~Yb|sL}3myoY~KYn|nR9sr=o
zq)y((BnG$3effHY^<2L?hUA_y3=aatLg9U=wcHH7YkRKtl|MVq+5c0kQlPU_s1EhY
zQ7e?G43LM|zxp=>3F?8CP5ln*@B;0%RLyxhGB#8y?N$?P*XXVER_ZENsOD2n`ISwu
zo05-&8b#jP5BiU*W2Ix(CiXf+4`Mwr6#Ee{Sg)w5c7<C<G*wWKM+n+CBZcZU!=3Dd
z_E{<+4|o0@4n&%TXFDE;kA;W2K45M;XM1hiEVM;yn^`r}cUVqxZTF@Tvt2dVAhA|%
zt)^HDTy4B>%ce;{D5qVuAB4{lU#T|8Phu^$By`xOQ@dkd`8Ez!K5o*v%iARTtHhH;
zD=aV6Kq^PJ0Sr_Fc!&MpKwaA$xwm%IWH6-ipQNtjX%r6DE9?2KD0k{Q%0Uc^#%kxJ
z<J6GA9b%lKb7;22AR1v~g5pD`Rcz&|qvF8>)G4r!zgB1=HP@wcwc>Uc?@>m(+G<&p
zB<G6W+@H~L#ocU_wA3{=yq>ya+#c+wO%>)uey1K=-^dF>9f>x?dJaH-x9)XZ5|Qxe
zPzATbyeBK$;-jX}{9s4wDwAkTiqv#iY@KMAU1u$Ej8a<q7KFDVN5vi1EZ5=SM(cW8
zJNrtmn;hWIq7Ny|mY@U#pLQP3cSTGH@<P3vHB)&9nIgNT7s1RBr0r5Z`p>I6Y9ooW
z_q2iLTIzXuuBX40Zd;SGN&eq|+=q_>AE?`?FVNgT?XW==)f+0UO^2683`9kWf;c=)
zJ*+%so9B!P8kl(dXC)gLmHbi(_-MO9er>E0x*4{ig~6?n!>o<c<Igq5*)7;IT)$`>
zvfpY%uMjcl-)JkPk#w1EhGt6l36h#^dxC8>u}Ihk(*!UPx&TCY+_sgJnJz>^QGe5K
zz6{ia{7U?d|6#9*tx?kDVrG%;2NE`J;x+_YSrfTN0T<ec9E3)|W1vX1PQK^_WE8LG
z1DNTCd)gPGDpT1S#Sv=*B!F6!A;D>6EUEG-%uQxLG2Fq5+xgbSXQeg~l5#Uqq7FKN
z{ArC>{9Y@#iM}4LtJF2-VMWqJ*eq(eP9AQJ#1FLmQ*fv7D0_~~LM_--WE{7#<hEf7
zXA}-nJrda>vu~UTS#M$etgv}5*NvGW)gTI>8{&8kBn;FAygISk@tQ>?gZRWMC^qDj
z9OsEL#h&{m-;7-e8}-W49=4EU^2arJOwJ0fidG|HxLd*~Ap@KQq*y-YD1cZ!zwD_K
zk%68!`e&wL=qg=*g2cXgj?+7l^~_o@Zg(5wV7Y}Shz?>N!@a3kek;2fb1=!sX1PJ2
zA>MJ*X(TrAF96$SVE>r<BL|t8u)>e`neE%LeQ0yIEfV94rQXZG<B!l~WOoDWd2BjY
zHW1_@%WaUr74^dZw%wqM2?(ua?iLsTmSDJ+UeFJ^0Fsy}{l}x^UrF7GxxxzQx%3~h
zQXYiBVn=>C4UmcID-X#`Gz}MT;<Lm{TsQ9$4~9MD0r5Z%TYyoyrmtvKO(9E#`ozD;
ze4-LI%2Qd52or*jWh>fwPqz8JJTGuVu4|jgE+Bh3e3FDd#!2E5F+jK67`8QWqOyXI
zW`_f(q)_H4ydAxmDB^JCeP9aLGa%?U2Rz6L*De>g@-KXsEz^Jt6F?fBLyS;bDx;Ne
z_*kwNd0aND!--}5R`533L$1RfvP}<%%@Fw#t4<!m=7~8)HMPouU;{mE6hbKvEbt6c
z&cM6jLi0h3j!Gq}0^7KSV09GcY9alS>rz#cM9(bzAUWD|fNxK@ZC3s{*jHYMPq7c@
zfAB`~5z)z5lHZs-kM;r{7%l@xsE<%>WFO90w!<}ov(OCU2AytBAj@gb;UmH^Ya@9K
zb8O=d>pM$zYz?`G8;c9VdMs+qmA?R#e2IN0x4?{2)3~F)3E*Y0j^Bec6!swF$pTvx
zIM4YYnH`N*OB!jYgkB(~lSb@H>;z;twhOOeg~-8JOj*60BW>OLk$7FRnC~6hS(XoR
z%4c*OyPfE1IgO{8B%-l(C$L!7=eG2Ypi9X|($Sp(b((sE%S*UJ`;@7^k4RAU(w+Pz
zYEIHD{vF+_V3st?x`-Tw-VtVt3&fLw(@BfjP+7%Z15Q>@rmL$HP6+n!X(|uD%)GTO
z;|b3ozAf3^`oXpcXc?>}RtFg6CN_d!h4mntl<{MuVJlXwxIv@(8K#A+!PQBXZKvR>
zNCV=9k|m86s<3HDMe&=xIp1&375rQo&KVnRY3dp%76%Ydw0V{c+g4%}v6ei6yyMR%
zmGUIATArimD7%ts98ZVAcJdf<Sq>1nj~w@|xA_BQ)Dp9kjdNU->mn1dcET-~F5?gj
zXpwD_{EWU@bdo&lI+o%wb+831lN2P&McptU)E=ISn~g(?BGL<`mC>JMq{jG5$%}kz
zD=)dS>j9(rm#Bey>er*~`7-2JmJ<dmRgjBr_Oj^t-1=O1>ml4m!tgeBoBf~(lp7Zu
z#hP&$)^2`6jNl99^@6AHiu|MC5uzfPEe(hjgQk9RoMncwn-CxmaeerCIK==$1LaDf
zlz)%(W*YG|$o|+FuDa3$oIvJ_y~w@EcA+}@72!yOjSxj7S8B;T7H_a)@naZ{8^tbS
zP4eW%XTnhQu6&B?XnUg_A?zU=-bnaeX7NsPk&yxygc{m+8Ty$n#>|nw^5;z5)h>Da
zBwQ^G1dtBwaw&`q;Y*dF(!b<mDrSohn_(4_uIhg!2HNIxI#2=}_^pgTZIT%SU-~WC
z896Hgm*`RAEU6lo?9VRON3KRxmO3yWO&9(1nL%tJR4Pq%o#d?6yLd&=?FWG$)*I%!
z$a1N<vdM6dza0SBsZtAc82n92Uq6ZPuuajdYW?}|Y+LSkygQu9Y$|O-=+NQl4xG>b
zuiP<nRnJQHHlC@}LlyvMWf>U6A7i>pGYOSUE4il6GTZWj_!1%$eA@VQac_Av+1kH9
zctg0WOoSdQX1b=jlsF={6gR;vxq`3E+Ki2Jowjc5A-=24QqUem=vIN>$x<;RI3Hgf
zY(!Kezv*R>l&>)Nfu%fUI2nv_y@6m<r%uhRglqwe)jhT&NbSTM*g2%C_X3)WdL;uh
z&-<Ja^fk0Ak{;QW|GQFKd6-1)oVIx#ktlEU1MDz%i-OBNY;(k6g2z>jaQQm#D3s?1
z6RBUs$?P*YkLYQfgI^{`5xvNGVWYg9>&%pQJwW~e3*^xS?L0$(r?lDhguKV?$7kqA
z@g#NyI4{Kn7a^;`nZiAFGZ3=Ppm!=c>~;7qQo`?-ZbA}pLF_6O2G;Po<Ze+1Ka}bP
zIN_OPj`)i_9=<~eY_76V?!yc>hoHWJC1OY%zBU7y6gXn-80ZJhw}_^pfeXkaPc0${
z&a3sEWxyT$54eYHk-nI270koV1Mkt#<S1)r#lEqJd=VOKIE#Mf74{zn+FH5RDC2yb
z6l+Qcn1e{Ba#7hSr|1Tv5PDMng6#07m>Lt0EkA%=WgP6qY^C50v%PSv?5B5#gALi_
zA`T?`5+4F<bG{<u<geR17}L<l;ASvcf4X1;*tpzD_6zVUl1gk*ao!xrEu2bRRL+VB
zYDIIXA$oAn6!j5*2`P}<mHofVVS!D-I|w>(rQCq4j#ihOW~-=)YiXN?|G?YgQJ%(r
z$NH$Dm@ceWE+71Zaqt8AdGJ<WU8^1VSE(jGD!Rd)l}-x7F(<G^{%#F%n9oC$CnR}Y
zK{?-zlAZwuI0+;}cZ*7-0n~%SDe}Ph0pbN=wbYn?2F)sX&dy9cOkRLbkZAsM`G<9K
zcGc)aY`hc%zXooJBY*|a4mhGKFBO3fa1B%yxB@OE=7|YZv#d+VG~7Y%L#+Hb>yPC*
zauzZQ-U<1HZ}?uglzeWTkKE@EYP<M$tj*fV+Qs~r|546g$Wo$f!8d}JU+_PYH=`wz
zh}A+ZY8>?!H%3j&=Y!4~CD0DK2U%6B4u0STBMggRDZN12E<Z9~2bbCJvW*kh6Js2m
zv}{`-XeIOI2KYAdi(nP2X={w@(5{}&f?#Q4-=AGq2<6C7J7hI7Q|*mZ2%b<X<%acN
zbJHEM#6#&Ip6OXmRI&C41{AC*7_5Jhd|GjY?+~Q&4bGrhxDa0})xcXOED<@>pe$E9
z@y*F}{yG9;2l=~57?Tv8WU`HxSMkR!6AQD%KGF#O6Ej8a?$^tg5D!_3Hu4TBScrF5
z7YX^~u7sV|y8Ke?mb3z0n)42iF~0^zGF428PKrbL*Fra@3AO^QNIsHVoBE54wG%mo
zJI^f_R*1c%izLGK#!d%b<Q+2L5Ke5CjrN7yPG4MwZYX4Igj5xuqogf`9EeO7P8C%_
zp2L-i{=$FCC({h^wyg3&&k*^(c`bg!Iux8Fe?d1DR+An4ESc4}GOp#vGP6hr);w^@
z+g%>P9FhkXHo<wkJ9;TD&v;e7tNvi}!CYh}>Q_1|MsO-ot+r(Q!`2ak1&8Ia`7T+4
z%uv1Zvgjpfl_c6*g9FUJn9Jf>nFQQK2<%F9<L=4DF)hU3to4LM<S_FDn;>lQenO1k
zKg{{$G@;PuWus!DB;gNA4=9zX=}Fx{tKVh0L9pm_>7jJnSE#I~(|lv_LHGn2WhxaI
z8+nlrl4v__t4ByusnMydlv^iOAxad2$j~g9%C{yv+g#a!U_O?dpMg!}w<nH4+VC#h
zN8-9tw@hP%<A<6a5K>}S%tL(4j>@xa@4@SQGwWeYry7Fix8PE9d<eH)I%9nsNY1Y=
zUzOg96P3O)89=!RnQLrrr&lcw4Hy2_y%R%3hBB94i|w(U<%80A-nX%9<fW|{{vAfa
zm10!>4f6y_ggybRV3Du+mXX07IV^|=UL$kysX|4u2agM_bqn$r7fF;s+3SrV=CPyr
z;pE9+d15W%6INsInIpVR)GzBjqXRXOO_E*dr2Hj2C6ca2#;b#b3UVp28N11#y0y}&
z1QRk7pC%XhH{&^l-Q-%7-53&oAb9a~0<a#iVueo9Ol;#D;JxThW)iJJwG+FbX|^3U
zuL;A?q89+YouqmBX?A;$A3YT~A#RsnN7@nV$~*#IQx>@c+8J0HGmyw+n`d`mI%4k;
z4l0({mf^C_Y)$JZe+4;N-%VZ|G}(gsvGRt5BW6@=X1vcg^<Ts~6Q#r#bPgID7-?yx
zb|x0eKg(w4Kga}A*#HHn6#T@z@CfpaCAM~p(8u6%{<H7~Eps}=Z=$aRol2bldoa_w
zh+2q^AdC1!@*&^A5?972ROkMb$4OH~s~Eqshg4ne1@1-qz^ARPkT;g5#_<$Sw#+u>
z)mU4koRnW%Mlnf-drEI_b$&hiL*8U@>4%sW;^{;PzbQ^2$h9imT+|5*fpW-W_A}@$
z!b7~+(TXWLS+boYe#$3oCuo{n4Iju3)2;YzcA|z;O03n?gF1RDYqs>Y4p1jbvE*xs
zuBGri(w7{UkQBVid{TZ|ApEp-vu&P#A#Y|!$rG?qTc7y)nhT^a>`?<w^i_`lA~XOV
zV_1L<LW_XGb^?E97^w`AH)n5R+Mw6g7TSwt7720*_#|>1Fq7U&`~}YzRvXfzFC;&*
z2K1n>(Vg5ed7;=$9Fsf)ZkEAf@!(-OS8<UNvIB1pq-<(0x0HSjyeP|#KIX+3XQ+>l
z6B7WQm|Npz`4L<-?iT#IOz-)`k1Ct(8>%Cfm&&SOQ<jMzj(=3Q;=RZNatEv1T(4S1
zZ6k9FS))>7?#fBEs@qmt+afw{7ua0hu5_0S$<>9e_!HA!KSE3p`+zHvwxpk!2nC_h
zY&B%CbW_j^gOh6qP8!DW&v7@M%Rf!tk3<-ow-Mn>UdmQRp5f*3T%lc=iqszX93RHh
zbHe!MBo~YOacr2he^X3>4Ic)U@B_Vz`Flc71m#J8wjPTqM=Zs5Wi*qGlm$J@9K%jC
zHe-2<C`05NUSyjk^`agDHhNA%|0G@E2=fr>cOQ(nvvcX`WeD=E2>QMJ2Jp8$2Vae5
ztV@a7;zZ*dw39duyvmypcUkAXk<^f=LAI3?ZwwX!^E?;UHO4d9F7zy}ztR%FpR`ZW
z^z)P$=qQkirt=5*;l61?Ci>K0o4@E|GRjM5_&#P<xhYN;cY<f>->`4eko@{|7wn>M
zk@Ab4gWCx}{K<^UMZK|^#vaN-z8yT!c!Ai9&s0)9Q;|K$KSV9HgLqRsojcJeif&?E
zLLw;pXmka9Ah{KhXtHoU`Feqc{;ElNXdM6AyHDa{ua)8UiSPyMm&6jFFY^gH=C$#C
z)IW(^$>zix+#7L{@YdYZvO^L4t&K;xAKn;drurp$3|!TViq8eyDh3XsARqFqB%VQ;
zB1P1}LvU$Ava;L~2d&CzW;&493ayLWfh&U<p82Amn*<*Qr|@5}9s&@U$_?3+7)T)|
z$r^VQ=8&GiJoGP5Jy;cQC8ofu)|FKTEQ_xgJizGj+n&osz4CW^`!TQbYFjqP9+i8G
zBm5e1RjP}MfPx=KE+l+3cjUJloAINO54;XNXKZXK)AN%?c$1WS!f&QNXsR?&ma&1p
z5kOFE0TWCq-^4ge-T+>}kb<eg7cauMEj%gYu$fq~)LEQnS_R(5kHGQqLs*xx0Zt^I
z1NX2-L@Q5K>_2S#`cCK!rA*B?bKErV20Gh2hU{)iW3TXI#5+<QOJixOd=rWj_W4K2
z`?#l;nPfX8hB2WR=v?%RP^xPiprG0?V>-cB=C66(QYGe_@t$a9gUC3(z2PYf$}gl!
zq7`h$M$ln|n6aJz#%&Wy=^gr2%BKqJ<zDC?$Voi};L<tqFTNW-tni0ufakFF#6{vR
zRj*G~T1vN#<Dk>XR&h0)4$bt(vV(+j+$`x8mLjDjG`6I(Z3MSYU`I*|5Vddw?iW6S
zSH$kTR#+E`M=O(3d_%c!!585Yzt6mY94j30zoa{vZZjKTDX}%-GxwI7an(sD))hU3
z!EzUInUIRch^HYZTcXSsV(31}1Bu<}29QU;0oSuJV-4O&4}guzql6ApF*r!Bj{aIU
z)2@^S+dI#B<!`<tV~_!pkAGzhL%$juk&8tuwwoS<g@xGw!uCTJ7Id(-LKDk+ct8Ik
zp-}0|bdq05Pl~SRd$50@f$&cxlGIk5%IkPHu!k*x3y3#JLqr4Y=u0?>nOe{T_${d|
zyBA-K_V;x`iv3oLE!*yWhiGUa`j9vV>7-H?$|CR|;9^6k!cKA<L|4>`c?NDrJDUb<
zxNoTgoj2kH%XWemdCxNa3sVAP(8BdbxD#?78I3JOPJ3&J3bq@ytr+6pzHU6&kNiq!
z`KKb41Shn}4;UAjAHq$5^(Y{e5&nSD7Y_7ihh}G6bWB@lESqoSm<wQUQy#R<)7=kL
zQISm3P_txFJT|Rq!9DLd)<!nBbkrTlw&a(`yF(e+AX5i_PhY60I#M0zFSe6Uz$0um
z5-+b@XZ@1TdZIoUpCL{Ip7G^2&>kp%emKKl(Zb*npvZg=A@p<W9LKs!iP|L)FZ@}(
zr^gAr+>~Z&%H`2H1qA<^{R~XhWf{{fea+i9G)J=GC(K>?l997C0mZHF!2|RR;0kcF
z;2-n9*n7SQT?=VPUyAR|)fT<r$Xq>qnO{W{L<exKr9PPE|2w`H_gB&iydO9TuWWe&
zrkT_5A^d@4A|Zg3D~t;FazB+t`QI9s(ZBhf`Wlb}{|E{2hD6%(gl%l}AwH$ZJT3bZ
zbP}|Bhq4zC5PV_&$;5!2ndX+L;DZ~KA?lvVGCD=7##}0zVQwp6mZrR$U!H#jhyahh
zaXuSY*LVxL?W+yu${28&dkb9x=Yku+PFWLVm`{-IabZjE?KSkDgrnkZ?g?=&xemIX
z>?GcSs_3$TEm#ixo9--H#Wqo>sILV+2P(@)X`eajnUB1oQ{>8GT+9T1HE@;pSRPWM
zy<)BD=KE-WO9{8WavMtu@{NDSzv8FDt-XEGg=~7!J}iY9WQOA^^P6(g(51my`NNP>
zxIX#_-^K`e$AvilH)4bLf*57Qm|33p{8S^FG!r<OT!&ZzUNOF;Pk<pX)$29A($6PM
zbg|I|<XejAE$l|<89K<*8$gBskUFTzd?lwoG9JBR80Jj?eLz;xAVZ3!1@UkE?c^+A
zyncT2FVgCy1?aZ?yBm&|afbEvYy3CR-|3E(JL?}KDFrjlGvO}YbV$#Nn1Q^`H3Q$X
z(+a<G%fU)~Gx|kg_3V#8wfqj?LeD;!!$@L`=@(`KnBu7bHcM(_iiaP9iC7Q#d%`xb
zKiChxgO320gLmkUNEL6&j!)p971O=_WjkJoT;iQb1rb9h=|7>rLa)I@^gMWqhZwcU
zm;DF!gj>AzFL{GBklk%b@gxd|aw9Mxi6&6c$#5RX&oqDYHR5o{g|0`W_$Ay1b}c_;
z^$))c1FX|Df)6D>AiBU0n9hL*$acV2Fp+rU+3021o0ttLB9B|1$gEHiY#_V?ON9eS
zGqk|}5w;t3$OwKa_)2<VnM;?Kn&sDn=Rlj+Bp_WRA5&Hj3pT(xbVq9uGi3V!Bu$cl
zacbj&QZ}1;f(%jY$(_*HY!RFb4*;{+la{@-)BKbz5UH~07O&fZ$j~I3Mtd2isrBPc
zMg55nWt`>UI6MD<tSvBrZg~ppfSM+=NI$kk{#V9aXknT`bEu{EGsa11t<7!GB*Exg
zjEoO7GB-dn;t$Z7{Bj{fKh)zRKNL>Do_W`yvzV8bMV8b4_i#_FR-OaiY5Ipn**?g1
ztdHjxv+Dg|$@N%S1NIx_M5CrS)<VvrR|&Pei@|$&z2Pw3mA{5Q;l}HOo*u$3vjI2)
z_rte98B7n|Kl~=k7vpaJAiR|)g{xp@@?Ex~v_`z=6~SZP-OM(m8`{IG(7QI3X)yeI
z_=5L`uRQt`aDw&FG=3k@mtT`}o2ky)fLb*+Adh%Iy$9LpzsVjmZ(;k;MQ9`GG}b7%
z2|C3&g;PL1wglDk>ISm#B$JyqSn7Ja7xYW6DzDUc72C@n`0B<DjD@RbjN$tT?fCKN
zEj&BA!BT+zG)k5L`VrejN4OOU`;i}U^U-8r7__FSjkh9?3KNZu=}tmRk8{;fLnnU-
ziRE|aw17|PH2SbNTL-aMv5RaW1|hk|1^k!<J3b-%CoC6!;b)kwg=_c~e7rb1aXY<;
zkIlG_*7G)wPl(ZRcVi2)6H(E8*!w)(2_92uM`_D-U#Y&?CKIeLd<xu>Dw<!K%L|#n
zKs`*K<X&4u#2$K-v=8dYusPF^J;rhMx2fy+p-6pimRN~BEA1$l&1dl&qz6b9;h+c3
zkHz=qi3KUXlZD%`{r>t)U97rqRJMbmq+JN_EiZS0o(rkaP2?DKtE{j0;<`Yc=mtd}
zdC2+|fs-f6z4MBJv_eY0gJ}S6K8G|lUavL>OolJAqxq}IK6*rcgloxLfg<lL2@+P)
zOG{^a>OqtEU4=2UR=CXH)prZ^`We1E<%TCD+$c4Rd)Y1|N!O$MXFH)o$#=kEWp=3<
zcrN{e9i6`z^xFmjLyal4$vc91YtBs?<7o;P7)F?H__uTA(KDuBg<E11FvGabUrycZ
zNy4(2_55^po&K$H1sx}30BtRI=`<P<?lZsQb`Ca)g|g6_YwI2Dt;)YGY>us`*K$_=
z4|X5iAKoKA@s#slg}TGPiVu)@=}&{=&o4vlHNXx9wa|*@KmnT*Dg2KuZ&@wq_$EnV
z;{dQWG#yDp_oBDC_@Xq^A=Ad>i_jCgCZEPLQJ8b)_bRgZUeU(nJn)wIU24qbGlTVA
z^=a%(%YI<0xiNS#Im_I#R!0u5^}|x=9p$UTFY(;UZwaNLZvV%mSr(bk@c>{Gaa75J
zr0J&YqN&i;_;lI=G_^d)6M*Z$b;KtceVouK@1^AxJ5!hse>NT_=)!8-`<F33j5$`g
z8B+^=EgR@B$-_MLjQeOk-OAj8e`!2n`OR30y)^TtsYq9!-~0m_AvP=AS(L&wlOx^`
zU!N^PP8&yIHOgCsay*}@7FFT@q6b0a!79E5*ebdPV@56J`9l5t`7v`sZtknOQ$c;p
zRsT^2EAwCe!xx*UGWF?R^tG5no?v_<{z1`Abdhm1@s_SDScqM|zFMjILtahzsL)9G
z4_&A?!sjhV`8eYleLMJgTt(o4<u4kr9?w}1UI*vE1CR&wG}N?l2R8`ap;rNtKZcHm
z*P*qsLU1tiR1SOUc`^J1UQu_vU~Pp&<5Ot7SBI_St=wub5gg8br(<){<wHa>_O86Y
zmP=j;Oa#_tk3pV`RUp{EF4<}N2u@=caxuPv9>2MiuqRDlUlmQZ{#4qcj|2)UVr-bJ
zte3XpbaLys9eP*JXMeS%-|+hd>7F#UHE>NoOn*a)0~U&pIH&2)Onm(#;v{yR8w2+*
zDmD9%7P`*#e{{P|&*2*4g1|FmqL6KDnKd1vq5W(_Bu5vwa++nWJYUG5+hXmQv77%k
znSJG;5y&s%FYHn5KHbyb*Ib%g$oyo|nWLVjQZl3CA^i(T@V{iWd_PL^Ti6*`>!d{D
zsc${mL|Bk~O?LrtnEm>0p3a*tk!`#e>2z+JCulhgG)G>s&y&06o<dP1$TZ8l&$bu3
zidFEL`8UkNq!Qmlpsoc*76~=T^8SbZWkv1ayy~eY2fxCy0Q-nJO|v(iFgFFqSuP;|
zaPI_gV|!i}S8VQBDDkV8!_Zr-JM^7DX=+ao<tOH?XZ1j7?qOXG|9vC_nIXh7dxcrd
zK9){SM^@-Umg&AhtSESteoPZ+nB^VZMfl`7u<|>`xELgrTw<i@nx+xn&BRRoSBVws
zs?#@*EW@(j4QrF~jL)Q{`7?nT=xV4Q)`PxK#&o>ftYZCzGrCBA8r0K!s_;-2pEs8t
z9_RG$#y)Z`wwwr~dm<m1N%$gQY~n2DQ~8!iCG?q)#wF9kxHCm9=+}shgv9|&YQfgJ
zFRNqREyPa@K&Ei#3aXZ2?Nz>!@Fo5-bXu8%9OYkOW6=t5tQ<x!_!gmk`D32D$t-Qb
zvyn-Bee5LvpMEH$pdF1l@JX=5W0W=_dQ$^-Io_~LKVZep(RoZEe1h)dpNYH|jX5{j
z3?AkWiwikT8tvVPHj1rEZZQ3n_A&d71zbbaAx=mH64wLW@>Rjk+kp-0MXWZw-PkWT
zPCm~!!jJQV_$WKs+o!O3c$oCu*U>VEUrjfL8qf>)K*4zKQ$c$CtBsq8-a_B_w#Y8#
zC36c{$M-~r!?l>#;NG|`^kQrv_$ui?Q35~YRF-sXl$%~Qm2ZPTizE33Gz1R^+vJ@~
zEc4uYdH~gtEws1rhrt_DEF8hFn*;nq=)Kw)evfQGC+Nqa3FerhrN9W!K;#K>8B60@
zntS43@oOeGx(u3SsziDW=f#VKf7EmV6#_MwxqN*s2-Zf{@=da?p%X1zg$t0w9QIt(
zouMZt??iHdZX9EoB(UsOq<7^S+*zZ<EEJCr#h#~pBa;>I@js01=!@~Y%6tJGu=19(
zmMh{?Xr&ZHXNsNkZ@>j;a{goFa^V}M1@IXDhS1<7N0eX4)!ig<X>e!pPs>PVsC<|$
zZ|<#(!%ESrNCcfl7b6Sl0mfN)H_vnSAGR8kE41g26prPp@-d(f5e>Jo3D84JPR$MY
zCajVr(R0$Xh?WI5w!D`r+`&wxRj$}PSa_xD0Ho!=2h-5;$S_X_G}%`Np2IC-w<Pre
zxq_FbzR3wnR638;llOB|h#a&6*AnaI!{I5G!A0GDJ($|oa~ns~i1#i1p6tu7j9x<z
zi|NdL<PKA+Kjdu=mcW~0FY%_L!Te?870VWc#uv~XbFVX%#4eWnD4nlMI?N654Dd`r
zE{UJ=D?tcyCuyDG5||CwC#M^~ZtB7RCd^2D<-McZ<J&<zV!FZQ`M+R0+B<2!xjZxx
z{={$d*+OAvH=JnhVR;M``nP)r^540k#!nUqYJ$807qVZZ&v0$%40sj!x1hXODl8E%
z<=;hbBz;62`Z4H3dPUMowCd*h9xr>-fG4-WMxwj@1#lWVPrS>&5Dx_Y^}p1QEMxXh
zR=NYtqxYijtS!b}@)l+X)I_<;O)hLv?i)G<`5VyrZ-UF&PGSRoQN>Qe3A8EnNB(%k
z$!zq@Ma?)5cd{%ot}^rlE)`sfO=rREV%`F^_kV+LCmd&SW|96Eu3Rh$iMjE@W$?Xm
zTyk&T1vQqI84R8)NHz0MsJ*GZf3mMGuz>$3pDG*&98BoX)YIp%Uy7nYdI3=|$a6cE
z2TsCoO^=z@*fMy2!CSbums?&Zb_tt@@<qdZKYgkA$mBsWTa5P#DwsbcVNAXF!{Ap#
zI(?gIUA7Y;ncI2Ob+3Wx%u}BqIRxEc`)ter9Lyi_Q#l%}UdG_Wtf+x3LofMz<o5^j
zfiw9@S<9f}!jE*hf~Cv|Q>mw^{A<il{xxVcR0iXcOF68tfc_i&+k2wOTkDx&7}wZ$
zGJciVP{`y5=8P(P#}of7?^e!e5ej@<syV@=3Xck>0-=PF;QXXG(ZzNLGEG~{P{1+z
zLgG$~C#eJcMH&+OwahWl!lLj;pjo;H=I!2I!179E&;@C2uHbu<Iajb1P6S#4jhIi|
z0RBh(c5iF2oF2e>duEhn=IrFK=UM(|sTbDW7%@Eb{^Y;W{XL6KM_Aa@J*ixFf$px}
z&z-aMH0$WEoSR*Y^%GkF)qP_;XOOLigMJ0EgBLeF&Z)xpFt<m)<~PZ>owXj!Gsmy(
z4z;AGaVzqd<lMC+`efiIy@mh73o`-XFr?@D^A}7N5EHP;gnC<+x!SKLt}vfA_weVN
zerEfG4he11eH%Nk*Zb-+SF5Jwoys0wy`%U(Z$&18l(J2cS*B0mXYjgbXzWL3spVhq
zNvJA21gbz+wk*VccqaQ%_8I-wk_7%IROK@)jeYHty78%&+u4~cuIuXm2%Uxcn110?
z3zzzgh6ns>-+Fj~kMv(Q!4^M!uUrAR#3%4h%V_u*Gzeae{AU~tWrNwCRcJXOp8sgt
zpWie0W^B8ni%@<26n`&&L$(UH0r=*9C$u%S@VzX!$4&QBU~}Y8pqH<X`9;<qLk2g-
zvX5_MTw%JEWTFpvpZk8$6MfC0-?$!pJTSpmA*oAQpB+e7E0E$R(|sWyIKjApdrR-5
z7beyB2pr)3k2?W>GjB08HtnG^=+De__-5=KZUwNJxkL->ZSh9}AcPqk{jSzr@P+@a
zxvOda`s|``UL}7^t_CuS?P6+ZSeV`3H#nKhKf;waM!m(ByKI)_H2acywQ)8$5jcTf
zwv6)h@jJ@-m{$4gpodHa?o?4_nDf^*-!N^29ocu8`uV$n3TQI;&|)dm&EK-W8UN(s
zEN$ow;w4_pt7Yhm%!A&5?SLJoe+x|J@xGkgi{J<zHJ5VbivH!#nCqBwe2e_;u#VOL
z5Ze^YU)O<Ok{j!JVV-7Ml6(X?RiuGy%c|)RPjkz1?@eQ<ax3B{+@p*!PvDMQL?Kzf
z+~elrQ|c$yD*4Y|sy>xh6y*w?U9Y8O(LT{mRBcyj3hd|(HV>zU4<tWVx;QhmuG9pp
zEuc5Jp>z3d(3sF058<Aa_rW#8^?OvOagL2{Tu#-hP<@qFrEMbB;E~Qec}DnrL2=pU
zX`wjSUMYNs>ZHWEKFhP6PlFOsPaH>ZTtuwPG_{?yFG2pWJ6v1w=i$?8o#Mx-t>mS4
ztE-i*h-mE8xg|<N7CJTrZ-wrMuJR)Sl;a*nq%?93b<7JV7USsf;Ko{iIFIu|@<Dh$
zGgQ$5E=#fFdFW`wEt{?B<iFY`cUtPH@FQ|@?dibD@I&er5mmnOtHAg4M)h^!SM^o&
zAJ^X)Z{HKz<~~YiZ(d6D5qGHL$(m{xN^|YuqvBOzp?7UKlX5yoMpjt|60O-9_JGO-
zdIqX$v96`~HS0EceHaULR_B)aiigN1cfd6zJTmmd_&1uuE|1Qk^6BE#Qv7O}M`n2R
zrjj0BR{faVFcN3Q0&SErktF$8D9SamF0mFwDyDvOzoZ8GFB4(MMu$b;4~c{uII@`8
z+B4+2AJVS#dyyuLNdCchmm9N!^GT>l*|z{fV%f&yOXpeZpUDF3wLX`6Aalbj=r^T)
z+a#@g+4jU2F@ZCwt)%(z5LcPD8EUV-x9@S7t$(}wC?WrzNcYH#@Gr_uIV+`uZMEtP
zepAK@yR^OJpVVe*vjetIa4ombcGL_Qqhsu}wKdYpvD&`enPL5s(!#MIFq}L~&z8>Q
zrbUvJjo{1h;D}w_pfq+ngI~f|!iR%TBB!}!)GJq8)X3M27>Pa+R?#f)g6)L;)DLA(
z=x=7P?6T%rr-$A}K1HxlRjL{BF6E7q75PJ%=E{N60(0veL0#<7(5UFPAY!ap9FW$?
z4WwNujp$>JTijrIcc6mfoZB545t?ehV8=>o0YScpqpfQ-`8;~kc_FB2m2KJhcQOkz
zI3Fu+A&i{R<Vb(VC$WpVPRzt$=O^Y5+p%Dxv&8zA1oY{V-Rk9F%Rp~32OsSkN;ZuC
z;@A>8LcWpKsH5bnNVe81;v)x9^N<Z<NA*|Xv2(OLwrnS&Mf9|{sZ!C!);{X~9Pt#N
z4PMn6B2Tp%-~)oE<~Z(#s#AYTKjnsQhqIf#o_%?^B2Z%c3ODesMbcA-X|;u+Zr*m7
zI_GMS4zsU?oc7V$8>#?mNT$|V!`_gdIt~UqfKxPG@j7Su$QCi4{OSHB3bIR`tEpK^
zY;YtxoLUzY*d1!wE`-BnUrE+QazcNR6Rmp34EP)}-?rPbOs(jck$O^_7(FYVDWv6=
z)G_~2jn-C&evQ-*Ey2o%+9YEsy#vdIQSg^Q{lI+Qr7eq`3hxZHbi^oQ>tqn?n3PCA
zmr={MXKHutGwtc66RkUKgB_8`?Vy3Qkn@1=|3}eThNrE*e|%eKUA@HuNeB=H+RokG
z-QC^Sxx3Yi6Ci;kBtSbob35B_dgkt~XYRi9&;MOtv=<j?o;!D*=kxvCzP-o_@7~}&
zt~*>yI!9CrZOPp4Z=f%vj$0O2AIM#ZE(1nkZ-SY^Z0l<HHTevV4<Czu6>i6Zsd%7_
zy3Njx&bMqdOVJ*9MY13LDWd`Ni`*JHY^~}&6Lpvm>%Ty8+@MHZw2I)NUotTJM66w!
zqZ92Ks$J}Q1SV>wRHt@jHS}(`rjdgqBzp|4j1Gl7nk|~O$a|{Dj|bY8B?YbQJN*Ld
zjo@TcniTdY(6doV{ga3qMhKMkYDtcHwef1n|Hw?LAh~7M?$9QgYaN5`HrlW|;@`nL
zoFDvRh%&v|@4}SuoY-U2-m<1yugS@_2zZ=WPF%7+cEjPO_=j+7sW^5EH4r)I&`6y5
zhV(z;-_SM3Y-)|6m3ar%o4*+QAzhM|qVuRc>5;j+lSplg%;p2)b?~v~dudAKu<)Ij
z6=N9+od}R-edrtdg|5w?HnEZ^I3v2&_>Z+hv|;HBZhh<%nHn5RjSEVVj(9vaJNv;#
z(usy6k=muZhzk7F*wgHbVpHNgTEDbRdWJrh*``G5b=m*Sb(uq1Vpbk|)B0VIB!m75
zS%^*c^yF??8|$m$4XmrkKNQ7<!bxaI&Q-M)FmoHCduCo?Pcjbg60^t3*dAN7VLjPP
z_Y_@;A9hWU3~XIuP4pJtJ}_6Dh>qtufU)s>v1y7U%=hqPmUrUx&0WZ4WHP0W^@pbg
zs-w;L1ma5Berv^GyFe=8F;>m&Mn9yMp=YH>^d902zC-FQcEP$yw&+kQ3v>Fzo)OmE
z&Ap-D#SFF=s};Fa<_Yfzl!MpAF6oQN9sCl@bHgsz#o!4xSu9sHB-TFqKKj#j8h?Ye
z*bMX8gwft6rZ;@X>Iq)ViJb^sw!9J4*fPemDLK3r>m2M99gLPUPsjJhYL-@s91*XQ
z7$b6Htu6E-`3^r6>}Rbiq)Cb9+S-FS<9Wgk<Q8&`ZB?xGyqIx3RS^B;^@R75*OAKn
z8lK2p6AfrTQ7YmU$yf(iV^ULbG|2crGoSd`u!C%h&SmOb?%^iOQ7l(ALQD%^3(P=%
zSoa87tQ8%M%`aPtJd90@c9QA}S>|s<!?F*hhoxTOW<(e{%|4=<=|OQN+b60C&57+I
z&qXr~-^^{LAlVerS<`||$${2<>+Ntdl16;xov0&1WmYW9C+i4%!=2<b=X0znu82&s
z=Z3n47ILl06;TW^`&SzdVMoLRbVsCzv7*4Rm5r6T)y8Bj&1&^`Ol3uD)_UVv%O+~D
zpekK}hH5@xLF8q`Nlnmgw<O^aQ?96BuSoaB)@EaATdZ~F6#QUxEY`Iwop<@}nm~UY
zykF?IG*{|x{%&5lzK}X!xRRV0aZ$C<d6DWh0L_A27@Z<akF5-^K#rP^GB?pN^0Q-!
zHAy;S{lUG6&E*_ISQup-LIZiXvU^8dawf6~H``jB8kO~41gy#auFeDGX0#_<Nz6uf
zMeO=Ed>8T!nWD`m+lVh!tLVE-VOm?$0mTl)Q#!&}n-qwjrO&Y(JURSYEatX}BlR1k
z0mMXWPV^?XFtRCp-Ta(=3yst{ESJqE?ECnmtTQ$leIZpU8z4#H$C(qvW=2~6mUr-j
zd`w&xu30)8%U69XE3$MJUT4-3?z?*98POsXaZj<{$x1+b@B{HN(rMI-Dg)WEdZrm=
z+n}V#De4-hu`Y_O5dT$9r83Cj`tyb;Ihc4xEDW=rYgC@;-&i+;3+h0f($#bvGLxlW
z(VN^P(++wibpdS?njmLtG9$Xk5p=Hj*5BCpiyOrMostk4Pi>7lqQl8q+|=k1L2Jnu
zpy)?xtW;6D9qow>COU+dh>3x)Ffp{zY?m_0M^cnr?&HkQxUT3%@r=@D-Oimezcr69
zD;FFi!p80q*wi7?z)V>hl`5GvCP3044Fv@rAA0R8Bx<=bEhh4k{<uiV^(r~o8GL2z
zfgU!`^48A+t$(PC*x1c!l)zSAL*NnWI}s;Y@lz@_yA{a|1!En^%XnONOXh;*gIJDK
zMXOSNIo;GP*2UP-tTiRzf7`Z_Qwe|RWotuIv*2<l)g@RqlayIauD6oZ8GM9%LZ!s+
z`o~%RKb75-q@uUbnZ+$Emx6nV&r(FVO^xuq=03;L!kzKv)q9oAwzl&g3RjAJi&U`e
zl}XdzP+X?ncoFSF_VM?$8l+a_zZ@4UkDtNn2>ay>%Rl4+tg+5%9b}%v%`Y1s38Xzm
z{;>`)EVHH%OUWioTda~+m)#}y6RAs-geC^T=zVLo*kiG?_90%IpB3&vdaa_hzNI~R
z72QMLg4<h)_}$i2vl7L)(}5^DA~ui3gj921%m0j~a5MAzNDf(C>K8g#dSs3_FD0qW
zQRpsDTYg}8yttKB3#@b%8-sSRDpc*ts%EwqR%CCm{=v3Wdh;eMn@F{cG@UTjE!B`E
zScuX`TtU*4PR!&=BYUh(V-1AmW=b4b+D#mq<&aK_ul4WJCL&EO?ZO{PKbOQ!jIPQw
zSz4Agr*cqxaMfnqu+TcwIy19<xN&fCpj)JJOo|<qJKL62oaG;KFE-0MNGfm8ONr4R
zp)dFml-9>d_Gnw+k$GgKGS@sZFgn}thngM(BpsiIj%GRtzfBuqw)C6DZ(hjHg6COV
zi-`rx%(KM7{0}kRy02`fvI)9}O~NONdqsr*YHbj>lbIKr7kd;b2oH?3!nTEO(17uD
z)|9eed@nApbgO@Rq(QVPHja4;H<I&PBdG<%m-JE6oG_PtBK83S&||VjsjYQ_xsQD1
zMxe>~XG-lDKx_}L7o*mL>~3Lha2U`qW>@avbIHfnXrLU@P+yrV31?6l+9}Z*{Nl0;
zkt60a!5g}mbrHIR1>EbXcBL!LBce;Wn}+AX>T<ex4)!4Piqs_HC%<OSLAS`ZfK^Bw
zw%)V`Y*u<lyaXKxFGxKK6^b>H_L4q(A=--D6Q&HSQ9iO$z^rYkIec88EYcA@K-J<t
zSQGfvtYcO&)-R&Lua}10xnYdii9Y22vThHL5|0{tF@vqsqjlEb6lz)DvW21U+ByD;
z!b$Zms=xg1q9=P)KV)u(dLtLi^$D9eF?u)pDF(?zKAed~ldVB?NHh)|A;IP~)(KoQ
z!}$<s|DzMdtpU`uovasmZ#|v0OlW}(#wJ;on9PxWK0fj`T<Z9q9WCt{eu1qr%nyOW
zjX>vcrLfU@!F&VV$6t#a%PbJX{78dNnuNs_?~u&W1pYkNiFy;+<QXeJgI-%!8e`=8
z(sERDaU9u0I;3eH4XPNvg+*X@t?o#j)+{ac6qmyb#3s@)VgordauVJsEU>ba%kYok
zdAO5l7hc|KEgWt-TlPsRA<wXQv>bZM-GdX+eb@vlEc7*W_o&P|U7YedT2uTN%!v#p
zr^$JmG5iXC0eLopliSgqvAXnf>=a^Vk}bbWsnR;q!`MospQQmZ-8duEuB?^et$50+
z6;sU>*#4GM;TuxZ@zA1$PY0g@|Kn>%Y|$d?ecwE=QLGocH2WX9kGI7Vi0b4rOD=QN
z@iPAf;tzB$E#kXTC5G9tOR-)29nBuNys4pOO5|_rbAga*g4M{*Y&Uox2x4E&;YcFE
z@KN6z@pg<1C&vzQ?J}2`IlK+FN%~HL(sXuP=?-Xq_Sx7gY?^q<^h^Zt%*@5~gxEvk
zW9g$<ypC~VW|4jv{28vDzCE0SY~o+DYf4{2)mb;6E;TSi)>`seT&2(w%eO#)fQYx|
zZdg}NE6u^GiLIfZOjmqj*(-n`CW|5C2zV6N0dEv-8M$g^(2eF}%wTZ`bDmlOmpC%D
zdUT2B5e@pL#m-=g&{CpjXcGAVs!BXTTS^_YF61-vEszGzw#?;*2&ttFq{9vtnMx)3
znnGteiAafV<KJn2dGq*Pk;>L5!WQFfT{^j(_$Hb7!Bz(%azE_Qrt!v6=q+k)pr(0w
z=((dk7ejh+&%?LHjpl4F2ayOb)jHfP_K(=tx(!o?>eD;JwWZTOlQG|#Ngj~;OJ|@O
zp}KG@&w4+@u4GYZkvJ8*M^rQaXTr>X$Pji~IECjjN^()~Jozbkb9iv1zU5(YZsx0K
zRqI;xh?HQ>4vi)+lKt>7c`>Y{elQhevS7R6OxS|{2PgRJ!dszw)MRO^{2p-!(?;x$
z`)JQdKMRZG2nFOa3bMqTu9AZd2JUjS2XPktf`NE5vb&sdT*dbn^7sLndT~88%lI>J
z)=_3vgj*7jV+Fn>RwXhGoxs!V2I^`I<LX7#$aa_{$70R-da5FUx30h)#>NH#ZwT6?
z2{l5dM~pN2URoG?4KEQJ#Lf_qB@~&D-3S-TPaBNrMz*zeHn#HLAZA40pyk37{8NLS
z&CfzFO(R5erd_++Tv=*F6-P3Iu}oa3SgP`U6o(?65pYv8e;2M;?4R8*@=hujo)}4u
zbmTg7$y8tK4>w_*N?Onrkx^p**tS5i=^Zg9Qju(K`mEk2H88H0Hj=5z=~Oo|52$Ev
zq64H)<P2>#T3-x=u9|B^vkW=`V(D-tLd}$k%|rL(th#{E+-%Vp&e?pITy55H$1GPu
z*;ud0J8_`}jC>@mu`PbR?8R!uo|F67DR^!yff-`PBUKDrOPfX7hnC<U35)bKnrnH^
zowN2u;J`|BYP7wyF7$U8;k9^G%Mo5hO%UqyX=D`HMKp|z;&);Ttbs^Rq|~y`Jf<{L
zcwmTxAIsU`9saLUNKQRoM?1hPO}zFW;yE(P{5-f@=m)c4PecEZzNoX_NA_gek@wC0
z@i|myu(m%Ct-zfG3alovx<i3K2acLYk{quVKN*HtjK1Zf+N3jFi(cc4$@Hkp-#UCy
zc6NQ%9x-N%_re=uwG3uoZEI5ayWZd)7Q2D0K$iHGL4q7*UFJ%Tt|tadehv)P$?Qc9
zR!%CYB6jr653Q%4O0Nw4@Z0(r0}@}vH^wp6>w?mf3mc3#$#${5&~&VGso<UwDvfsH
z;$lkZ$BJoS3f$H9TwmMA(a9N0d`c(knFg$aziL_{IhxAwA<rZ4DSx?j<w{_GQ+-3t
zVC_K|_Q<4ndLA<k?hHLg07FGh&_%)TT&s&}!emC0+`U=oX}VCxGAPM;*F8bo4Cr8A
zE%W*t(-PI~G~0ZiJp+6vpceK|u3qXj;8CL=NC4B-C*Xsi5<Z@Pyg;u#5AISeaCL-B
zT{YALd|kEUY&`VdF<GOu35BV;CaH*aTDdB@@15!P6Y83Df6s%AY-LsL?&Ra1HncV|
z-4^lK;&i}^)NY1EThQb1SKruJ`4JkH{#;j}ADi4<6JPGQ!UJ?is-@noRE@QI`*~|&
z0pBZc4MQB=TOYIo&=yS|-BSZWQ=vDux}J_qPu1VrMb&O4cLLHAryD;2gkur25xA_h
zL&a$=3))wGtofO~(l6MK`vw$g6&N@v{c-v$yN*8Mc&mxJa%i<@c1AohAaq(cUfBW)
z(38AlisI7zzPjFWy1~BN`SsPb7o|@Y&C?dqR{9jA1U5szK*m#4GRJ$WuoXOW`FL;*
z*frNy2x?k;w%f-s>x;8oNBu9HoO*noo}RCsMmGXBFYgV)@`+SUWdvftB+nN(fpLNg
zMuD`^XuNZrX0Rjt2Vk8k0F^!-IuIBJ3a;k4Dex$mcg^xuQWt}bnFwuiZ?JDo$%p>i
zIMm%(b2v~*C%C5gCjhvY_qV5;C_f}TE(oegWiMLHTbDl#a22=}T}rz8HTijp;mB%u
z*G3*{$*`_L-ug0cFaZ{{iBL7;0^bHY4}1>vV%FGdDfOPC8O>I|^be<>+81Vw2d>){
znm16K(WQ8<+2lT1?DwYS-&2p*A<hvQSzf2Q^{V;GVD2)g0kkAAG`&1?vTzD`!<Fc>
z1d5b%Rv*txLjKU}H?&e$vS-0|+U)zE?z`uKBE@~6+8EU*?V;Rrs&u;4wu$!Xa=-za
z21yN}vw(&@riuX;@2hH`=%w0AKzy~$n$v+&U|28{{s`<!4XGajTV35!Jgx!{0uXux
z+2wq_?ulxbdxq|q_o}<E9&|kP#jF4JlFq-Xd`{`)EO7kvT(a-<)^j}a%yV_t{9Eaj
ze!hN<x0`l}XBqg+8>jxy-_SSHHHS_D&!+4uQEtrAjzvy^sevW=yOgh7wSdRkVLBA(
zm^v7IoF4((0R7ySigUpU8u`wGs_EFmwgj%-?G=HU%CXwo^clqq#Ww$HXhyOUss`4|
zcnVIJDg12#i?5BdeY&aWP&C;a)OB~KIA$j8&qH<psNJ5c@Cf=1o##C5KL@Ue8*iWC
zSLxU5RM1Y<N_)1WN1?%2Sgi#;O0mdoRll^U>}ejQopHBub@M*eZ%;RQJb|@v#Fc2I
zTtAr<)hlh1M+4?Clu80as#XB&X$%f^mujy2t^;KmPFEjrjXSC-_P(~61A~=zx`ykb
zZn6C_y$HDD@+k6|-SMA;Gr?uW-4k;FGn-M=sKP#aNipp|oN<PppnKwK;ZAUT_iY8M
zt6arSeGvBAN`VP!scgTD>2{~?Zq=^EKODV$n}NFazDZkMcQXz_Q{9Jj@9gW9Lwx6a
zLtObjujX&$rNf>uH)Fc4hih`iDgBVZ)WRd+ZTB?Sra+qGQeIO!OJCXZR>L|xU=8Pf
z-GXWj+}VX270V0t8b&)n3HjmlA8@=AuYA;X)A_i#hi09Aow`<XEH&g(7c%rF{|8r4
zaZTqgsYJ(n_UK=@b>P+HK=DRsysC%#mbN{#Q?V;O+1Dtev8F|!4vqN+&`WHt_~kTQ
zSf;&hU!&enZ(jFD-4O2WHh>519GJljQXSgRn(nAAv0ZW3OIq!23(fQ01ZeM1JD;8?
zXDMQ~N%kwn%RP>wWzb&5l9HYYFg(Igz!vA_s785y7A|(RMmYD1z!23)<sQchU5<L4
zvsZd8n+B}w`{r#8?#-W9^2yf4)}2wZ|N2S+e)(u|d&yGZinby!Ib|h0E%k!m=Uxtc
z2dbx7)qao!`zLtlTDEv~l5&@)&~-pJQn^yIC1X3V$#noI@9O27OZRg(^`7FWKri5s
z7P9~FZL;^#GT<(-M7`3x81(BtD!YJ}J*x`0FqE5KFBfrJmf~JfN#KRHI@kox(f+Hw
z2-kt?dar}m?9aVhlegLzYQX%<^n6z%nL6r9XQ(r@YwgDsHWjRF&Hi*e(n9_rYA%J)
z5aI=y5NhFOJ>iU8`k{9h^FK{_|L(jox_DoeH6C?sMMzs-XLCASb#%B~U*GD+d&fK8
zc&mH+fM*Lk`~ULI^t1qd3Z?8`Ty1>m9|D%zCwYrpuk-`dHP)G(w-x_r+sB)jl}Sre
zO$mE+IVHm@#!A*|0q}|Y2E1Fjn!X7B@!*N2Hqdv#+sd=TvYE~+=vEk171OV@eu#BF
z1SU9Z`bMYsrXQtF3K!E|G;YHbK%ss_*HUoH4iszs|GJmK@!B)+FtD#qP8|bV^E;;L
zU2i?39kzn_M9_{vf9!MJ60Lz_&eN_$_hH>RSFQ>Pw)gIIZg3v;)^}A@Bm%!wS-y}e
zqOGn0(m%Qa^ftp$&n!rx8R9)uJkHfZd(C}MUZHH!<kSY~V{LY1gl?qooG#N<L;p;n
z^tCIO;vc3O?e~MLy}vc4w8>yc&41pP%Bh~>#ER-W-kS7Wl^dE2{H2(y%>mOZMzy_t
zvk>{(X!CTjRF!{%cZQs8pAJ4`nycD)jzb%?6MY%%B~N3<V|$rf)3%`KG}2nJu27$G
z*jqc}QQ=+H_Oy@M$*v{Eo7@Ya4p6h?X(?}AkIHXycGO+bJ@y#Lb|rJPTRijOKenhV
zS6v55R(;n7Y?lifYcxLE4HgV>9@9MxTq%ECb5`wf%`^ZY=Rd8h=$#c`U*8UJd4rDn
zx@IXS%P;UPcif~OX|-BN)f{MG@8{_YElAnqU%3VVMif<dXJ`r-QZc|ck3O&a=$r-3
za{N`i-7}QFo;RINSDO`=p?4`^*GS(yUvA!jlr-9J|D;$L$klHK8_~JCD$ES`D^(Zp
zBAu^o;+&UKnVzXRC=&q<?4#%c#Z*m4sAr%;0EhRh`-BRd4((&)GrcKq7_vwWlx%@s
zdRw|N-#Mtech<@-t^%D2I+=LNp>~WYY@oR7ZoBS-{jhJgwpMO8V1PXid8zxPX|39t
zR>Qm6=S=xvTcC3rPPopd*0HIa%|X6UmmG9}t}xtK3p-D+gm)n`6z)vVrSJF}I{Jbw
z6x@c(YkLHu`V4i?k}4&&V9>Wi^$AK~cDQ<I^R&Mc0`N?=!ar3zoUQ`@QEvhpFgZ4*
zn^i-a<AwJsnr%6{6z6HrhP2^<_u%67uT%}`_u8?J4)89$4Va+%+Z$(~3gZ+lH9+BZ
zSgq{iYp31po&Y9z#s!WirFq7veyGg)lldxejQlz8wHGu47_(1VSPOn%aTYx?y)QVW
z_!|V$pWI&)#%ij$4mwi6+~Rk|E^jS&71wCz47mbj*}mj`c8=6eNi6{Sd#Y##>R^4o
z^DLI)>$<WHWY9KBJ??XZYWkK^sFv$Bc;~y<+YTj8$dBbVE7mF|rwvwqw;ywd0<D-d
zW&>Ovd|B?3w_|E6XaKFx-{a2)W@uD&2U}dKPS&-Sx*NC->MH0*xPJl4yb^WClJ6T9
z(cA1fnjwCU-dJI>FG@dmMArkLA2QQ-oP;~*&Cw*-dTFcCRq`7$7l4i04Zi*UQSPda
zE{;u5d$}81x4c=}$^gx5rqB7ZHIEeOz;u6q@PDqs@C`CxI|z@|ywFJC$Up)*F!-OE
z@~loC;K@io3>}ewrz5!EIZ4?PG=pt)Pmx7nn~b9wBa{a;pFNQKat5v$oqT}yl+<>e
zaGcPf_K}*Zp7WX-9wi*lpH`^V$DtExOMHmtzm%jjnRjp|c(*u@(DO<rf_uTHOwICZ
zT)owh|53?G7omR+%~MS&uI}mOirVTZcDvFPMZTHfTmNH3t2wVHHq?P$x^v*Kt}E(o
zntXo~U1wmo^NRi-|D@D6`efY%aFsUf*pu?g^Vs=6MSX99p$`}d;Pf26QM%D~nXcy6
z71b`x1<u+e&l=xzdp-BefJ{~aW9iQNN5E*!Zbz!FADs)=rjIE5dTz3<G+X_h>B{yI
zzD0^fippS_Z@r&$or8zc9sR$+oTP=zI};Y)FVzBPf4Cm&c1{K+uFKROrW0tNrc47s
zKYab{It1~3x3yljaD_r&v*@@=^w<CZ=m%TD+a5L39_BQweT9j~GBiHH=>_NcU}i%u
zyk5@PmOura0;#kzzQ+1K#h2Y5=t@9;?^0!+LZ-x^0w0lFsvV_mnz~Dopco1)^Dsq!
z!5!57Qs(<mM?FB$Zq|GSHzg6k3+TK0sJFnA=ql1wVn3y>_T2~puAy=_HOkv6F_@C$
z%0yVrV&%$}*VgXyiNJ9%KXJOgsjpJe;-m*@8plExivH{R17{>}@mJR*Xr>kW06I0^
zRXgRP_JO{)yBOHy1K=T!DR8wwf@?K+9xm@6rkO-Hqt7Y+gEpsDRo>FyH%#|!Q4FN3
zCO`Hz(slBy=_<~CfS3Ij7@7V<KQ+0XJ4+FCT`ruey{cW}oRV_R)zd}z^Xw~1Kzci1
z1y7|N)3kuA>w?Nw+ScGidqu}fm#Z+r<#3f>SGVYd_D0eM&j$rkxW-ope5&PBJInWf
zoGK3P>4cz1zDv&j8Pim_cL0F6HWeN7KX7mL*OwDzt+ZM6PVIoSJB9bO*|rvi_X6$V
zo9+(UG!MoOfr}kxDI0i7ue06po>8>P?UPoq;6F!b-7inN;tIIRl|(P{5Xx*@4an+e
z{aw=nfjvM|r^O!x8hR?FFLGXU73=PS_iY2$A6G8}E<;~6H`LeDQBOxrea}F;CLQP9
zLwAS1(_^3n7f=6{7KK7cu6Kz48|$+#1zX$u(TR@L^b6S$o2S9jbeet87jU0uyrZ1|
zbLyqSAF7$!kM34JIYq6ju~rQiyIZ5DJ+9<h$vb?9bPc`Qe7j3hoYM?YzECW7r2;jf
z(e^&f3U>=`iYveJeZ?d7LjQ8@N#JZ+Nc&DVjjlq!Pu$~c<k{w4pxq5-JL))+eDmOQ
z?xVJ&iv8-Xxd$|<d23bs-2>!v!=}(XO_H*f<}XDSuwOe1XrRUEbsM_cGTi-B+`0&O
zg|6>i<C#VOty-<x>LJ+?#UH#4ypKE>@ZNt;eW7S4FbI$hJ#feo^i)pI_D*$NPIcKH
z+H!0#*vPdJs-es&d7sz8o#i}dF9%N1D3i5bN&_d_mFB`uPIK`+ZB6wOWm=U~6$ot5
zW_XS|Dm&VP^X(rC%QU8pXW&-XhN72rMbB%6zgSYAfyT&0cdm1uTnnGOW~Elm&j8=0
zU)D7Ck1m|*QWV`(K1xFrchl;C(^l0}Iut?sWcvc;+zh}z%ylhb*81tAnwEARF40W)
zjdmZk-*gP~wANg-#j7}tS^LArx_<i795|h*odlMs=DXK;Ca7L&Q1@1_oyOwJ&R?aR
zp&hNl5_4_!d<iMT-IfhQ)$M^hK<$D?Mg3G)Jp#}tF}rG3#shk`re8_B^ltRUq`|I*
z%Fo&{`VOw)U@Bb#q^d`{;$^CFx=czwQ11tO+Yc4B^Qv9(j$-#@XA^aI|0igB@iFE2
z@&LUu11hPlSPvEe(ej15gWwI|uC|fp6IfeLq-?5i14#DGazD1^*q5q$dcN4!BU?3t
za;fFjO4<W3LOI%2eO#Qb{99#k$APOA&s;98$t7qiyAP^2d%T5xbclEB`nDx!Z2i3J
z(xz)Gs{I9B@{Rfysc&4{U76|MTy-5USC|~#fy*%JO3|Dxb`&06cW2}6qNVmr?#Vi%
zdr_`>r4iKG?mG{FCEiu8WAeBD9(O~x$#p_A$F()NGGa?;?e1akRT9uH_q6rNuh^~^
z`3F<_GT-xFD=~LN-F)>)@P#HOuPhr_FGZ&|_!I@!yla};w#|ym8}2Epd#ge9;|H#u
zsotd9t=tfIAKaw52EDA%CwVcv$TNptkpEk~Q5$tDZT*V2E7s6ON?+0YJW7`>^W?RF
zHIBim4X(ag5p-x*+1I-|m!v8;=jPM<mH*KGqIWXQd&tuYEX(_l9Hc*Lj^*R_wQ)P$
zcNM$neAiCrDOGbfY&SWUs6g-QlB@1*n$a4$XLj|bTi7sditVHNR#i+JXQv%al0)&B
z`>xXH$tm8V?Bl(dXi%3+1ASwh6E$lT<21)~KE*pvSy7JXZ}&o1OW$L4L#Q{91iV;Q
z2YhXBmC=-r(!*U(k~Vpa>678Qz(-9x-7feFFe&ejj--2N|IknDb6mA`$Mo%2T}_>;
z`JyqTR&iH`g0{Xub=NZ61-rq&JQcEkw6RHF9R(%7eWR<o99|$NBSClEI~cM%{;Igc
zb;23knB;t+{*eg#N_{#ySG!NeD&utuWZuR~YE#l5`i`n5__69*)h%s7dTqxR#chuf
zJ_v!Da*8j-7W*W<E`3zeO&#RvtH0!I57h>LyU)S9*(DY`F_8U$EasEN0@58$$!HLO
zg#O4HiLkDa#!`K`!QrtntNsnvG_odqaa}*dShA>ekdxFO<ei>=<Q!~D?3U;x>zD^{
zTCBFl%0DLOL^O^|_;&OG_D!m&n~JK%%CT$shuBLlB)lq`$}bfE;l2o~h4I>UOsPD{
zK)rbgIR(x$n~}->Z^FuG6{Xi-f?m=r#Pj)~?6}Bg<5g*jz9YSeP7@76GkpW=W2}t5
z$^Ok&!L~s&XrAgKp`a}|!a0|ps*I(#iCMUs=r{4UcD1>loM->U_Misy5n~S6j2%h!
zB|$VHT7hVTeKkSQPLvP7E~7I0lOL(ep{1FBLPNL${xIn>SgBE{jW2-a;}5bH!4!N7
zU&1y)%adKew$@!ngVY{<TH?pSaAm3>mPFMh58+4Ue9L~J4Bz5^Wae}&Nh|ahygyn|
z_>T*5Yxy7O?#v&IS36K#%R5YOGavZp$QHp9Zj-8A@TGB>zKrzA`SKeD5snhA<PUO5
zW+^w*{uoU(JtbkweDNp#t5j>~%r=p2tJjbTL@C!bX0SdCuMwJX|N13UeR~{tmk1M0
zq!ROg<TsQiFZ*}pI>L_(GtB2maaH?39D4}YkfYhDCBMb0Oh>Xr8kr{=FQuL2%E)U%
zC-DMC8EvGB+zcK^NDYk5H&#NEO+xqvnFa|$b#p3tBm~D68&~nQVttT1{6*%V;kLCz
zBuXNgaoCmUpvYux4eJfd^YAm%+2B0rE@vbu;*Z&=%ZY9yb3~)7Gj$Am6XC3fj9>Un
zqh$EO$_^X$FOxz%fMpopm0hs(iPYv_VBzpJw3h|qFGnxp7H&MIV15wejM+Frw1y#M
zOw|hFR{lP@5r2ZFU<<KPd|&Z-+6d(`qZNO~1Tj~{k6MM2;3jfCS{3~hzRCBB-i)3K
z)MU>~t@YEGz6EQA%h4g8X8cMNm+Hiha~Uy^-!9FfZU^6+=2>~o8~CIF;66qLbgZf}
z*4*-ynJRa-m$~II82;cMhYi=$p=MB+`wygvhUQPPTJ9)0*VG5O?*GSUmmbsk>gkk-
zuVNQqojkWFfVhm`GnTnabwz=3A-go1YsFPF{*jrrd3e0Lh50?YB=R&eB-k99VExLS
z4Gbj48yAQd@hygv!F$4lV1}g8?B|!89vU*D`sy2G+Iyi`Ld>CN3bj$1s%au17R`*b
z6u$_Nct~E!&82f8kQ>Bp5(isK6>UAQ$U9<Vp%S%*U5C+;_`p5G9`DOgfwW2l=+=>o
z;za#SvI^0Rco2Ap{1l4<w<HZXmsw?+WbJLakM-qy`w7=S;u>!=?o(hJsEjmRcRb1%
zKZrUXXwZX?ic_UAyb$^aeJJ})ej6ICw+M|bS<$Ar#v0H3r=$FJOwEjQ#02QLfi?7F
zi(N;}+mR;G<(a$jm0T6>T==oE8A}P7*j%g{*F2oA|6gd8;jGjg+ZE~sm1EcG-wF4z
z@kk}pc6D)}A5oKko8FZCE&8IrBP*!YWj#gM;t-z&zdEmr0r41tfnS1s!beSA0<!~G
zWczRjbZ@LXpP{O1nkCbhm+;e;xsocFE=uxzl&|zPyhLgpAOR?(VoX$eq#H6EZ5-|^
zb|)7I)7a<uh0J&bWPItWWtpolW>1(N%4d21TJ~|{k*f3-k~dMz%uq+yO)i7XFjf#6
zgz^Fh@dV>YugmnwIz6BOXSw>zj>p!Vk~nX;87?x_B)>rukfBK5sO&9a7DIoqEYIYq
zXjG9ygt5Mzu?*oWNRk8}pIzwJ3iZgU`Nye>XoAoe!+gn>RoF3dHa|9Y+E609zT&h^
z!voM+#Bl3u{x&}eJ*to2Ttn#0XMye5YEt)&HN5erg>0+P7S3F<QELiU7K*s1M7Vg7
zh?vVsS-d=}2N{pl77rP+_%q(?*bm7VnyJ~#Trf6AY~)M{g5_MdZY=vRK9YPHSiuRQ
zFFptTju=T5aS6y1*D%t@UFG|-%fy>OSw0Enm#1Ci*3-~cpWS#}6r~vo4^ac3Kmjs6
z+>XyfIteGN9hjd&LuN1D9qmk3r5ck3KzF{K@Pdej=6f@+k=X%iW^^6)1@jvd6>sQx
zDbbna{zEEIr&JWWAX_!#W2ZbX$ra%qQVV?n0fi^xQ&6P1{$JBA6ZiqVir1P>GnrDV
zT!j<3M%XVzL+nD&qbj<cza4fSZDlw`&11FUGtlYKUhin>H+i4xz+RAjJNpcY)HbXO
zwNIL8NOj$_UNYv8oyd~RW@3r4J)`0enEoj1aZQcod^Nbrx|&>fW;qKP&tp%VHLY(M
zkP2oSF=Z@l+3(vb{xR*S-GHenpH8k9l%WaHnuex9nDVA>#Uim|vURI@fJ0|ezll^+
z7vwD0G5izXk9oySTxalEa2@GJ?7p>VIo^@K(zlDsgu7Ac#40S;{vSHV-rK)PBt=>n
zF1~WT#m`e8!g%Zoo`Ngb-RM8&fnqr%-QHK668ej+V5|`I2v@l<?ll~QtB`-tUuLhk
zJysMxiR`47$ba$2-7SO|G}2Y3BGp`)TCC(Rb0KUf+fiKZn-RWmxCPz_uSKgB0$d05
zN8r2VC|5I>&5!Z-#pm)pLh{rU!&|Th77Onr7jezd!^mKNF*+Psl=TojP9~#oh={)>
zu0sYy{xkip-6`x~R*RR&Z1)vmi@qx#GQZ{Pq9^?Ojq$<}_iS@Zwk}d4C19U56;X))
zNsVBZM_z;^{<8IsFU*b)RO8!3n(%j!NobmX1UfJDfw)7pkp2zCP%*MxDhk`pqe8z0
zSdZ7JinI|Uu#$WsE|)EuJAe$Sks(B`79CjC=z8?7Wwj|uT4}Bwu7ghJqQoX(3^53M
zCbp6pHyd1xOhNkktA>YKzfe~>1e^u0jr<1|ho<XOEh2v^8bp7JJyKVoQ{a=Ku8|+Y
zBJ(Bm7#gqcqe?dgxEcOh7J!&zhDFNr5FHen%g&d5470`R>LA|2GDBF){z2aeNs-<r
zy>Fp;r8k3Ht5+Kr%f?(cVn%OBHFPg%Bx<s3mXw)&Ii=Xb4Mr0AIr1!kH~MPPBZ0Ya
zTjHf&N8T2yBRYO^=<ne5;NI{T{vmgjn~z=4lwkd&#Taa;5Ur>!Pahzz@n0;cbwl`-
zREv6tPP4|M0SxALqDfS`w3dH_n56ntYvUH_KC}=WO05nrcO1{ENY9Qep+<_ch^V!-
zyl3CSG}8Q)d@0q;h>;_>RpJzKsctb<nRA;5V4uY?=q%?4ZX9wDU7}Y<o}>1NnK)&7
zOD;A4;b~KaSS|jpbl;pty~Q7*Qwgh~3((aL>StOi)8(Ye<bBIAv<b5sABLUd&lw7U
zHvIhH9O9|ag~dogt{9C2kI+}KRVZXUj*k@oBld@j6^W5IvV+ZtwlMee9aeW0-r~hF
zfme)9wA|*GlBc2{QoEs5EcxVl!~4Wv;9_WiRKTAV5fBdzLU*tN`wB+L3BLE#V&aTp
zTO><1>Q6zQ5s$(uQ@W`ywL{m|To9RVF2#ByLfSTRjwM%kCZ1AO*7xwlqZ{~!89KBD
zxeC{bAz_bsSilp$ge_ziaiiQT$e4N%F^-B!kRAxt#MC7U=O&>%XDW9OTTITj$OZvE
zNSBg*$l7*?YpHU-=@m6J)}Aa3w-22RotL^u_sOVO6&o*XvrGo}u<eAU@IIh5sW%GY
z0<vXxha!1bSkB}YGxLxSavkeQC@CSGgQl`N?t0)C;|do^&;3k9zFO!=qzShhZEPqE
zUJy*oDY2KNH6074m{B~@nr=8M%Col6n6#2?6WJKPqEU;D5Xm^mm}L%f>e$#wPHd^r
zR@@MNO%6n_!>_rJf1$Br<N-Gq`a3*<xbN*8du}|SOvHD{YwD$Rf~X(8YFQ5lB16R8
zq{bzJZIE5zN9upYHsU-Y(R;wMOIjS>!3KyFst>U&c0Q<;x>GfT0--lITN;7oioemt
z%sh8Fav#xD@dT~yY)b4zy9HGi7iEVA>JJAWKt<}2SR=A5(vTSzn1Z>o3b?`KEa{KD
z?pOF*8WW;%QQ3+v(}~TpFe=Ho(o-Wd*_$BUp^m6daZy``&@rehp1_==?e40^I#O4^
zlUyJ9j4ee!@c-i_OMSxyc~;gS(}?Ioe4c4{C``Y?hV$c0fU&Dsep3bPhH#x4ZF<J6
zh`OBZ;o3kgY&p6r@hQ$gxw?r34Y8Wm2ILH}r+E>X8@|Sj(AN?l8BRxFd<|AtdM4D3
zH53Vd1HK|U)^iW3W;wEKjWEx1OKf6IWRltH;$LK^lHTs~l3=JOwX(d$`V%9W{%Nfu
zX-E>hfJz~EGqs3ifyEqX3ao1H(_nd=LX7jQCwP1x*9F@~eX<mbDw*o~STO<bh3>X6
z;t6sNyI)Q`)%4Z?SzxiW-OygR#g~u%PJYWT1oYdS)`{$Kj*dh_C#;0~IlrD*AP$FS
zaEA?5g->V&DI{^!8N*PxXXgLVkf4r@j%6Z~LKM-{Iv>m7_OJ;-Ti|~Qlld&D4RIx0
z%2abMLlAK#Ucvs@@K>k>*95Ou5>I>x?K3tb9e5t!4tZ!(GYy0DkR&3(N3siSx9~)n
zMCPTz_~BRy|47af9U~vnHKodKn^@%EXI>mOGQ4&fxrDEZ3*0iKeZ-4SEh~rk_)M{P
z<gS>_OXxK5fp`?}$_?bc5<sXk(Iog8IU_Bh4j|_wCh3k;-D)?s4yB~eLvPqWll47G
z=wMPyv<eLhM-VM_gdN3ACdLzUGq<~L!nM&(*89?;a0hA{RNp+v^GkD^8V}7fd?Th&
z)BM}9!^XSPH14|L7}+Cug1YT$$mKvVH7syJY-emAnJOK_o60V)TI#!E3v(r@IrD{c
zlzC{6I14dqdUCMvggee{LJ-0m8qcxe`anabruSFWD;Abr4=3p~sv%tG@B?ZD*b<m7
zB%^!fm2u3nE#gD~CV_0qK1H(4?a5l4Tj=TQfZq*2tfmbQAq4X#rgG^8)`*?J$D1pK
zdSdl}p=1Wu1Gvv^BZ@pnr60B-%u%${=5xjdyv>*IduVDH92M<?PA3M3zlYnYyKq&C
z^8E+V+TI*zfBYq|4BJo~3w5S5WY^9PaSHSR-6y67R{9?6w)1f`PPTOe)PT@X`Y18p
zl!P}yEi6SW*TDK0is87LobUY)Z)EDPEi#^#dk99bN(*$AeGl;&$Z`D)c_smbZ3{OL
zyRjH1&j;fxptCfu+^wat_yIydZ-xd|=;_?(ejFMd`YS13u3{q%3V(lg1`zR&Afn*C
z@O<ZA+<UGNt$=rvjrk-|!Eltu#X_P?J0$qobPdVSbi|qn*CLp+VU?G7oRFe<U$~B)
zE_PE76&#rC_eaZNqjL9&VAw)6Oj^Je5Kp)t#<N@-;{xx$$XB>O7jNp~_oPwQ09ntQ
zN<T9hp}JUujS)JYdnUVSGqEolvx%36=3IZ7BI{zh0aD1_pcLM~&p;-Gx)@K|z6C70
zMeGXGPvbcHC$mgn&#;W@L~Ko49Qxb4(w{8Pp4uy0H4KQyvmvy{G=N#irLgLtZS^JL
zxVH_@Bvt25F$tzY#=E9-o=e6g{0BKod|*0`Vxda@e*z$-3SZ<Rp$DM>raH(+{(g9#
zeGs<HW!6f%x9<Dgag+cqf(hh!U(3KuEG04pd|)UGz7{%Af4FIwll;xV;uWDT@qkN3
z58z62ob2>#8oB^|N5<)^aGiAmIyhtmNAla@Hm)({M{vK{7`Mi8!vsU+pksJ<<bm-9
z7C|l=aOev+D%3%)IC@;g{ez^^$AuH<HB&!{E_aZ<Of6#O+Ul5Y8@dE9+f7IZ{Eu;g
zVISW`AdD4Er-ftTHoTMUij6d62+z<|e|h_7=9wv<=!maz`#e{DFW6FWy=-LM&Mq@p
z!~~<BZ$fPf_tAFBIPJ?JJBB)$?(r){Rj8fzf3hujs!2~;#MzRL%@@|o<XRzffM0`F
z6M*DzT#3eMKaVu@o`fcfL-;gfwZKw*GcX-!<C#E=VCs@Je1GXKhvcn%Gl&cz9~vex
zUa>YZE%d=Qi7FQ!C>=py@MKu4X=wjO-a^+yA)N>E8TjH4^lslT;Uzzpph7jsIMHUC
zf#zW;`s+v^^fsprU#A;}ni*S=HA0v(gM02?W2z(Xb8I&dzN2tSU_Iq#X!jO>Z|-xb
z5x*XONk_bT@N+0TumI^|x?=ppv#37Y7u+W}WCG$WdBf{;>q2d~TcLU3Dze>88(M0(
zqU(;lbba@<b~lIO2o+V=e-DYNp@1@Y+TWU-9;_srIo!G&WVCZV)l;?<ljs@pkA5*`
z#GfF8h%WpvB#B!{PtiQb8cMr}z1kYKt=JV)bF!iFo$(G=KQuY?3j4{8#+Fmf$t|Wt
zM<Ujp+Ndk3+Sm9RyvJ4|gZNyw5?)a`B<(ak2bHS{uEDoh=AadX%A$pSK-9+|nqaGl
zeekDjOq{Ie@Y9B~9>K5{%XS4bhjA|Ki%%0=;~g0F;`g0hiDc{+J|4mF+Ci1%!#87u
zznE$5x?>t==`TNP_F=#I3fL)MeQ#8@k5(hcV*)=UILL536fB*_ry)H|Ke#%^zM)ar
zH10JyQv2Od-DB|WX4>IPjJWoWu?hEQOGww9-5Tg_o~xTxfz6lw-DDUkBI7qsU`@eW
z%zXSTg0jW2f6>)sJ?aj=$1<7)_}ju&GET}tyNJErTJETK!R9he9M+2J2u^}qV-?F^
zt^1moEG2E;jI|2f4Sfsx3^m~SP(P+AQbBvc^P8WeU+7(gT+x3;w*W!;?en~ADwAVK
zcAJsJ-FLN0ZoF%^X}$ZGzAH7E`y8x|USvk1pXB}5C&qy5IJ{MuYPwGP(4fa|{M(;z
z4Eg$kspd84V04!8u3<Fr7tu^Or=NpX!dD@^ShIH}DTQAkF`_#ETJtZugIc>W2Sw0S
z@L2dP)J^Y{K%2sc`R}58Lch6H_{NN9=nbX>j0-n6Wb^5$mQ~XO>_>21Is`pumcdkL
zj-ed?KY9kel+G}@gH_pNe+%DYa;<U`ri<~|NA?Ex+%z8t(G69HMYhY;V4HBjbbJq_
za|x5h^1>uy3p*KaVtp4PvEgz``Vi5o!VGT$9wmjK1?|gh0%BY*XfM7d_&>A<8xWp`
z%}j=Ln&~6-fbVABvx0K7{*;)bxy^P`zC{46jPD=L4xckn%tdq&AK`}KJDGz}BmF_)
zD|?v$O34k437c^^)(v~d)hpgAA>0abCisLODE<fc_e<echT%w7#v;T*okgGMyD?4m
zX~?F)UEOA5LuzK~7>^kpfmdNfwg??al1v9f93M2ywYFk5d4j|#VxKl5dr}X>5%9gQ
z6EuO$K_2MUXdBHg{g-e8QXfB4f;$`HJtRcGoXaC;a--?fYejsal7uUf-~5H><sfYw
zP5#Z#M-~W;nfhS@n;2S5_k`M-zMK5sW~OEAV`L9DpM0jB#EHg_%yi_S#;fXSs!cQ}
z&lt1ds%R~&Gd7f%gf{yBt`z^s4cIq<m{nMT$%$?!HS!eIhiD-5KkS{h8ur3d9nB@a
z=^(y9w<|QzKa~jZ7j}ZtwNksF3M}(~5XSkp7<6)GqcszV(C(+9m}bj59hXelup{VS
z!Pbl;G)OGf)%GQ$3hoz_!tV@qky9=5T@yP%ffz^Aa&LLYeKj~w{%=+%d#dM1@q7%J
zBOEg}Vo9_jI*zm8V-0;wO@&_CULp)l^WT+jLUv*i{K7cLv|4Bq5JLnr6Ba@TGXL>i
z+OrHh&pDC1r5m-A#U{bNSjFfQ=m_y#PH3y~RmR4}>y?MNd-}<t)qFfvJDg5;T+su@
zxg@ft@?ES7(hJC?aIvwtk{;sLS<iC4kaVOHyWdzJ*`hBAPWLwSHSt0|lIWXty}AXN
zPI=ZPQUIrLoG|T)Ha49@$M_HDx%iX9U?<Az%t3D)@tu21{3Toh|EK-#%fk>V&O;a{
zR`l>If|up`kV&VTnin@Ry-!JEdz1f~hT;HXW51X+8K*e`j2f>hOymw0WV)l%`3rn1
zV-G{xrG^2<&7=j9LJ0kay9iuYbP9GQP6y60Elkb)SJ`4dN9^p_W0c*qJc`E~hcg;+
zn9+jI59MQ1i^nj><moS8u%-CakT>tT(1KnZnh=tZLD)DH2(92+5#^Xpp=+ii;je)h
zQ5tQ5-H<M#F?J6(f^Z0Lw6%2Uu|wu8V>R7dWFX&&U9GiA^K>|GkeW#+4L2BA*e3h`
zE^?QG6kbNP+2TM;$$ox|{x)`*Y!XUD<^!e1NnmZN24v6|GtGe)w(8+txu1iRxk9>!
ze*u!i7GTN3{Xm-Wjr%Bbk$r7g%FW3h6r4`p#_k(4eG`#$WEDTFCb^2j)D5$^{o*fS
zpP)CrFn9#Wb;GdUbr*Hh2(yw|?t6v#3(qjiwzkCknEIfzP3OFzezWnA>Z0Z;zLw86
z1)(*9-*`W?E+BCyvA4!8n<_KgLD)9P{~Mo(X&S#QYr<y|OH%FP+~AdPJo2w_OtDU~
z;j}=poW26tF0_kTWh?UDCMq(2LXF5$?*(W=%4~5lq2``RI5t}xV_1iGfQE<manJmX
zWee*Ex|*c77J{u1lQ<=D7r0R0U#dwh)&o#ubfw{i?F7;anIQPs&F(Ee(sxQ?c%`w1
z^98?D*zcdU>z9kc`iJ$wvB6iN8RBDh2eF*1#g5Cmj8%4z+0ljzhsTf)f{Q{IkoBky
z8iAiDWS=~d9jqR#5S-%Qn&~h+)WwCJ%uKomL{vKLctX4n-Z%F3<Z~O*k;=?)TWKd+
z$rNCFiV!%B(%5^6AK2DNvav^C4PVO;4}9P<xL%@_%|jdT?Ti|(O85$!pg78o<(tED
z+=&~*C4~F>XCl2}TpmT=Kg8O$nFfg*QD1rnuhSL-%fvF-$AqGNp*7G3dMo6UZOyIt
zp}uaap@}56UsVk|OOWJjF^yfzNitZY(L4E$+*5Yfw(jbVs$sHAd>B8Gu4Gy(ScD#Y
zLBU{Cefj6<=rMH)*aCqN*!ZDZL*a<Jj|OEWdYwE`P{B6SB@sXUERl?6h>XjPUf?(H
zh!p)YgrR-<TAFczN2y1F5?u@M2QmUnlIHu)hU0<=I!|iM|6{Cze4*ajB75G3I|pl<
zT>LiUA*4?9kgpv1)<0CeIXr_oh@R&Da}EP9`0`1cX*>4L_=s6;jFSiU6u}4mI$=e)
z=CZ}izg$n0qCXn&Rc*v6!RgxG!Rqpr3?p7-9aoQB>WML9@ZR!U&U8(QZU2_L;ePrS
zg+_g3S!PIUo=T?(i*$BAk$uJ9FCvhqu4&v|dlgb<tI^?}IH4bvm)fPQE><zLL6^w)
z3Vw!vcovv6UOb2wIW#KcG;lk#1RKT`P@{c^(G{k9=tMMu)sju)Q}{{9XG2H$Wv~Id
zOFJ9$%4Q6J&4`+V%l-A_U8N`BPyaag9JB-a+I1dJ;Om4Y!#hn3*Aps4j)=|7iNP9?
z2EhL{bf#fdRcjnKjLA@t@o?DZ?6vn^d!Kz66!nUsR$xl0q(G=i4yBS;lnBL~9uzGN
zhnR#jaHc)1wa-3);zUxGb`vz^xuGG350YVubu%#w2lU?V{=UAf^<lm5yZ-C{|IJl@
zB#%;mi4Wlklnj>Pf|>{54|1O1LZs41NV*^5t<I&9%e3m9=6k@I)Il!8dmFBjgXM3;
zyA5Na!r`=lGpLoyi+}3sl#aOnrrMyE_30}C^VOH}nW#&;9OIUR9`>g60&H}V+y|y-
z8hxeGv5#0->rh^3YBQB7q52_lB~{7J6njBBSS|U1HgAhkE-&JyC_7*~AkkB*7h1fn
z((n9Vf>CGz-TH2MgOVyG)`$2U!gIKkrr2zdc`z&LKg>MeCBdH?;mS4*2@<mcwb79z
z$4&;#M207b7)ltBW(Zkwo%;>Ag(>Aal?fCtt>t%_x`bPvwZd&p8_~!R_GEUdQs!Jk
z2GhgLer0j|XCkY6WJ`bt{YR7KQZt;vui%R52jX9rOdwO~Y$pByKk(?)Q&=xtle+LQ
zshA)y#1*OG%pXyoB3_Qd{q>uuS<WtDj7f9LdFzBGWe`nQr#g!u50@ASn;*Fq&1lH;
z1W=vKXr>!ku#sBMhEBM_IryE%WA3rxtz?JE0lR%Mc>Ixt@Rm{>pGAe}-*E5p-F3OB
z67E~3oK39RfR&0sc_lv=Ie<x<r#^GOZXdzwVqOvasAu9CbFgp|m3S@O`J-9hy_(7%
z0@}26s(!ADnx-Vl#lkoEAJJ(<Df5WAgWIaxos&{`fpqCCUue!@Rx9g-ZJI8D>jLRT
zUPR4oSkC+xWA>V<%(BVX$Mg^uHb*)D66q4hMt%q9QE!<}qG)P78NqMUXEFp^R2#1(
zn#UURr6F!3^PX><di>Z5Jej#-O_K`wbE-GCM;WgKp)u-4d_YOml%~iipLwKfunCHR
zL^0}R3zZ@FX8K-y9I~T+=k8G9a4)}|nK7c5K7;2dAH!u{i`1`-to3@PpQ!M?<#v<1
zl~qh7{1qkP6trH90gbl*5i_+;;ZO2?`j(YNi(r6S&NNZ0<agPYs`KvW*zdH86k7j<
z@pUmt9aJ`W_u$pe$7HB)n>Z4C*{?yHbv2krOsCIN+t@zkHKK_Uon<J{JjwkEjwL>C
z4uRu1x4GCF!qt*X#PgbyC|+~aR`H^|9Om<>=Jf^}^BH&ryc&O5%vM+0qlr|oQJIM1
zymJp_F}E0h<%N)^{Fd<R>Vw*sOo8qK%PJv<&4eH5IVBi&D7Df=Q(eSD=NB=#>TF*N
zf8XWjO(BA*Q_3hyyJxi+N;6y&n?z?bDe8RUU0*x<825wY+z>Z~ba6qz6t<Mjp#Q~n
zk?&4e&{XJKAeEuxY_h62&IrBgS><or0nJxi%1_b!?CoeNzg8HzJH<SxOcU$yM5#l%
zql8BWqYQaHu+b$TnfFk7LpM3c+-aKPnrHvCMo>C%K;&#|tu9NsU@w$fXbZj=(_x%i
zp38ou&G(nZ!`zq7yA+Lfh%oLjoM?Zm+K<@Z^f4;odvO^m=kvLAWh1{kmf=Ey&3V?f
zLRz1&(D^|01M37s;!@;)><QoMTcS)iHQ-&5dxa1tMblIFdx5(_EhY=#L35k=iLKW&
zAcQJw*l{Q7w3NIEbWW@kF#J<}i-=$*fs1k5sWiiFP#q7LdokIH2ix5N@*U%c#@qxw
z3ec#CUTEgM`W==rrX9xTC4JP1=1LSyRgv{f5FO!cQndeevY{f8zNzzr$skO*gJZa_
znNB8vIa!^9JyN5}hz?Vnv7b!^qq#(>fYz~VokcpgoGzq72S3X+S9F8lAgw;drPz!1
z;;nX(eW%gFHSvOyX3BKh>&1p2gc4>!!&&r_P9gwbz(+)PxQnPrX{R?A{Y1`k)R;aZ
zj#|Ix?nzgvLS`SI50_F`$Z4)}Y64zdtwXc+RVb6<u84W`2KzP1&;AU2CwAB_!qbKx
zG?ngW!?o<cwV)jj8Q#&jj{`(5n-Q69_tb~NK_t6+JjwKRP^rb1)1EKIFUcgS#+j#^
zRx8so^R}@Mtu!aWG;s>E){&^6`#gc$x%r+E?nYOR`DJ*>O^(e6!(q$yL2{kvXX^c^
zpTt*%_cXRoam0LKx^CQlPHeT_HSz2l=;ye_hWk>N;XIm5<*0!%$<c-EF?KZyaD2if
zjcu;JTlWC|AZ;i7g%~`IJ#E|K?KT7{2S8u==f-})h_AS|h|}Qs`bJk)ea)va=$2eV
zR;h>4A?Z}~KOKPUK@V+LrF3?UTaWKWXVIzXNY%->nCCacImBZe?nt1!3=H=L`=+<U
zDlq%C>X6<rEXCm?vFX5!Zm=&|T8)qG6kP?|M|H79M80#K@Ses_+#8z=Ps<0TD4gI}
z6Pyk%lZAX?T{$@<Yy@FZFUm`Ozj97}TKIt2g&s@S$O5Ak-z0YuVJNH8q21lD(f=l1
z^jM^;(%YsL>ND`6#P}7ltI3I)%DRwA1L@Weae#3X{e&pd?QwC`BQRfoo1z^C`Jo(1
zt)Ui+<Jga7Np%QkgcWKLeGTP`H0Y*V!9AEK1_3?dBQKM&_Uy>lsbu>@@1od0;1<^@
z>!rx=9Us+guAX3&>o+TZN+X;BY$X~)8)9E(zoOqa&Wx1AOmc+yEj<-Az)nKY|H2K}
z`~(gM%H6TQ&{ar#n}S7-?K+L3MY-y)%ej>&ajM}>Sg8v`k4J2BH)=Z$5vm!0iqL$c
zjmTsV6N#}`#7gBFP}nH2hg-?L<IK|-N#R6`#%*#bgCV=b8}+50Qco2t#+|U2qdGJj
zP1ep|!gcgwsn6c7C_(@;#Z(Hn30kk}xd8i66FJrUBNIf0h9`Ua<jo!erAU#n_i=`2
zD}BTj0+tDLkQF@>uaPG492kb9YtM`8gmARhGq-jR9v^cak7VvpH)8xLjP`jlq&Dg|
z=WKMJ2u8W62QBh78v?uwq*3<nsA}V8bj<n)Y<4dcb@D}xO|pa;iwzYwV<!q>;xl`n
zxE^KFJ+3^a4E+s;@mT4)bw4PAPq;&b1I-~T&`ixYdXalg><~7(kASDnnaqWn0oIQg
zc7{mLu>&B1ItmBS1zBTY!LM)^72)>xLewdqc3)zy!WUtwG2iH`K2FF&uyELzqHc|l
z@p|H%wsMCq3(@#@ew)48J5nxl=HpC|BaNy^WSYrnwgAn637%!JRQi?r#5oSMI0sA#
T)E-5F#{)T;RLg`{kxltO4xJu-

literal 0
HcmV?d00001

diff --git a/wasted/disco_test/single.npy b/wasted/disco_test/single.npy
new file mode 100644
index 0000000000000000000000000000000000000000..4561c4f04f1d75b08858dbbb53ac806bea70b170
GIT binary patch
literal 35968
zcmbSy^?wxC_dc${EkxIM<c{ua1b2cK_u^2XxI00L6_*e~60*CqyCZjOB?3W;7kg8n
zP>Qs;Yq0`_Z$AIRcYd2YKit>6?m727=XuVvy=VU(y#|*rH@Dm(=h$(x#>{j!GdWv~
zXPv<096xR5?3tscj+i!c?7071?=))itZ`-QvnGt1KCbK?VSp(Wm{@lTjRMmW)Bnd)
z!=coTEKD4vZDtnXYg~!yxDc%ugA2L2krbw$G>JQw^1*DSkCV?sQ&OAAHJKmbbzxIz
zEb~qp4Qnu?@y~E7(Ult%(Kv)Y!rTdwQNKGLlo~JnQ$yX@cFeD!Da92JDH0=y&@%Lv
z9?70yS1WtCyilbQiCf1V3B6SY$bUq-6wz7?+`u%5j@^BaYeI8eU#A}($V~!G*}tVn
zOgO18_s|848WH8H?Mf}L0OqS}B{j-jKV-m(u0xE6p68yT4o+eG_rXSe(DoMmD%VT5
z3Vdo(%j3rQ;C1{=dY$kkTG$(dQbrK@PHPa(j+91@r^FgtBWEGaW`c5@X@8n>PFRyN
zRZkCJbq`3(iu_v)=<2W^F)EVnoJ$+_vB*`re98mGuZ@dzO5t61jrZoAKyzPC^OG0c
zTfNno^)N!&m1m(UpcgftO+;@M*5z9oe^);qnD1y#UnDlBw{_3pj<c8a&+hus&#p#}
zyij@?5}vKDjV2@GBe&R(^!NT+?8sD6ZEF7znWOLEI!o!)E`DiPW<G`fQ(DrMLJcu1
zsGLSnZ1jhg4VZJBo-juUnCo&S?n7axdKGN3|HQVz>gqbir#x|;&gs*933m)uFCGPF
zN9uva-s_P+ltP1xo(d+=i}h_#N4kc1?5sS*JXsKIci=(y6FJ|uIC@6CL@pxtL`=-9
zI**uEdZQMV;YrSG{HJe-bc}q436forLl(D_?RVfRw@h4!3GQ&whFyPJrb~t7Ugjiw
zOl_@yF_uL;!UrWt%0wzhc}>+}Vj?E+ACXEHF|R@;^v~|0E+X~0`#pFOJ>njP&f{W3
zG??SYT|#s{H`Xo5!?>a9=|E5)FRm*c9A?7Dh~m)Vls}6;gkO{<hvpYO4|(-r^sbZ?
zLrUppOw%{wY1}4vhW1U{PCwLRxNF=uV;j-UH;ev}vx{Eq45FR3D*m)O5_S~hnIWzj
zp;eI;bhS_=9W$OpIMavd25t~tV4>ZM8xDkn+^Zw8I8RJrr#a-v7Plx>V^<k9y*bb>
z-X&W^?-Wg7{-S>)r?JzqEsjYPS^S@;mUuie)Z=n(jWjlMAeJ>n>(YO~>-uVHgkB-^
zG<ek2A#$DU9Q~1LWSof<vum8*vV41hrV-6tC=HZ0oR)eINjAE3M~&D1LU7kP0H^2=
zp_1b6hi^y8l*6w0@axF&kXMUpR=f&0M9gGW(?0byB1idX*U&L-W#E`e76S(#3L^?P
z>T`j;kYk<ERiRzn&%s5}Tcpe$h-5b3psz@&>8^l`k$)_#9IBBw*TtvoO8Paj0;>`k
z6guoOxRC91xV(BSPFGu`OiAyg%y%#18bmDm|I!9?V->mCqtI@)t~R>J$_ebBk;FZx
zVFe<?Z&A8ZHm*s^W!NyS7<^N*$=cv7dr_#s{>$;?1U>#R=UQ?gto#_R<tm@LpR@WW
zz#m;hLx;Cq3GT$t`y%0$;;6KRk;CD6;hC<v%!a}uM;%}H)LxObVHta5EGp5W>*&W|
zp6e*`SgXx`r5YM29LUt;Zl<NFtFis?FD@bclQD)k52mx#T;sv?!YARug$dyzVB%!i
zZCq#4saozy^xA^k<W=KT<WHij7Y*;!rgDA5v!b_=EXOLeHZvb&BM;d7DP0UKr5xUr
zj`|L;H6z~{>x?_$PwEtI8@-Vp7$%wH!ScQfDIMV2x{t~0;e1$)9WD*#o+(M(E@n*;
z$F`LZ@7c>tb(x%B%A%wgp5<2ROVw(Hl&7^;o@nmRhzzE?7~?{_L3Z&7ZZK0m^m7qO
zg@}Bn5|GFYY^^Od^`G!7S8uL+eth~7{#!U0S`ayo{0n|{ZYBp8Jw`_}U33@sf@wmG
zGKSL_yN(;CFSD*jBl|wMPKL+ZGQ()-iAW^PNTWfjF$RB4nb~XBD#*EL4gcmq*Jv|P
zhm6&)629W|<X5?6sID}*a0wes^KyUc{Vdbe%glMS3VBY>P)~%K!5R@jzKs6vQjq_V
zSa^>?$>;F@!YhkRktAcA+QeOgz0nSbF?K53MN>?RxDq7?n@#_&oGNwEi!Hs#g|55K
zTlj`(27V{7-gTNwNdKW&&-xT?8}%D=Qr*N3`V)P_&+AuM72aSD8xLW?=%VI17jkJ#
zD5Y=N&avuXWqo80-8~v|jpf*=J$+tqw(ep^DPQQR+%e}tZd1@(cqsgD(O|9uJV6(S
ztK)AY--ohOCb{FqVb+vL73HUhO-eCz;tbtM|DYb?u90S9R<W5aAw;<tG(`RC*R*mV
zGkn1{1pds-cM&Nk{j_3GQ}ykUU1?RQqtW~PKbB3=e@GhX+x!LFl{?4nXFBNBto4g?
zj2Cb?wUAv$Q$&T-<LRHU$*ymFuVx-sU)QsP-$i65SGL)PauGvPcQOYfiD5sp7FnzI
zC<O8~atGanp2BWGbKK|2X<^0KssE;4WIq})hB>m3u4rGT*1~`Dr;5#OlUr)VueiFd
zB!6wmhtI{w?%Py2-&jCqyZ?pTqPIZ@$a#~?BE3OqwzH~m7FmNoPRNlZ{4R|PUw!>S
zxsZmok%kq&jT~?%b16zM<aD?Y4^k~7(-~KEVf42^Yj&$v&#l6p#rx<~t~Gb7a0C5^
zXEwHnp2!XjJ(M(e`TT`?1>&3W4|xt*QM^S*$dNjx&v3P-pHc{PG9#G^N}cd8)FQ4%
zNiDe;y7a!rb+C+V=uVT%MJKo)P!GlLbdq|<{1AFdrYmdncGADm>y6*(Sodz46@P`v
zp~B$9=t84Hv?sR$u1u{|7;mg+=cNv%ySv`F(>O8GgLY6hwqj|Ps6rJ~2$u?rkpV4c
z>LbI;QlHUbhI7P#1<ZY-Yh<$PwYD@=*?r$0l^Yfw0o~~TG3~j3;b1bz#;|{en(6^I
zUtKJ3<X$taTsN3*kuORo<4iaY{gLeyEw&FT92cD=7aAm%8{J)2aWt^Ti!m3&or;<;
z1G$+6Z($Z|VKS*g{jxf^=puW`Xz#+^|2iC8hm!uZd0#CAW%5Hy!CREob{AXS&!cW8
zy|@}zqRrDEC#ud5%8AhTXt!`(ZMi&_P1}T#ufjE=(+YnCzp&$@ErhSIuU^i+%E*Un
z-Hvcu`cC%MP&ArD^oica8jyY2nknn;u1F$x0eP*K<skE#yBW5pWg?}%i7){0$b9ia
zt%~gNg5o_l_>a!fz6C~WgQHGI7@U*;i>wZ(Nryp6cq;jdy+D1U?}R&3F6yN?!MY80
zHlUHnUKN_=9SiUD-(hdk!`MdbXy$O3H+uUk!d6`O6ujtd$sEbUc$7uVR?=IT?TLgl
zkX1^qEu|FF2jQB+qiJF6XylGj73R_z)}!bj;7D4J-TvsGw$T&_a|*^S)GM)%L1UMO
zSI4p<%cK1o-XNP4w1_U$$EJ11Uf~1b@zennE*w_^^ytX)z`01e_(fHbl8~V+-#d^Q
zqHUF?7KdCF@$u-j(Acy=AzNW>z;PpM%rEqtY|3q6DswB;H1d1Yg}-!vh^#bPhuSGi
zpvn9)^>}!A6dITVPx-8uxQc}^`&FMO{wSsyGmUGYeRKdgCH`$zp<uLN&#|l6*S_0q
zZGE^rUA&CUVcv?hqbKZ*{i|)quupUi@}}#atD$s&xs&!GazkE0r=&Cze+Vbi`y*Xl
zKkI~H<Hu9Oh%2G~rh4wG?olG*TF3^A5w)88aJW*)<%%O7I3MD(Q>>9~+EL@OyE-z*
zxuIw;J;?Pf)PcQM)S4N?ywJ-ts#cHlYd&&$(c{#s)GNdhaYajo=|$1V#qeGKAbm}!
zYO~!~?cgJPBfUMmllhun4OD~)WPfgKYDf3)?i%5dSWnl=)JFK>=4K<qv61e<yq5RS
zx;(SU9UhNv)c**c3)e2Z$JFLx-CL1kp{|~<`blL52nnM@j1kX%<)(tR?lZ-YjRNW;
zU08UUtxx@=OQEr<=AKT3jADJJU{P6f>VB)<39lY$p7Kt$fj(Sn^g_Wps(s2Htv+|q
z<tkf0=XME1lyD7H?z=3;NbYa_4lzrc#w^g5IMVbEhdOcdclXh2)AL}MnaF9$k0L#p
zd3Z=^zW=iD!Sw@uFYQRELGo2oVaf#SPSz3I&2@!s7h0hy@=`g4EyoTi{65q=Lb;y1
zz8d|xHB5Wm0en!^N8n9T!r%vOH}jsfs-0QRJd-wrOxLaOH+EF?!M+l*7HJaP9O?g^
zX=`Mn<+(nX%hj80*t{9OJ5n_=Bx$g*t>gxMP~8z}f{juK6_ya~h0pj~IVT)(@XYvd
zqoT~nz@qY6UHS?eOEn<JMsN8C@#oaS&<Dn3d=V-}J_Rmk|BD?`PY1h-jfnwRygrT`
zn)cXer~MEui(CBb!Q#|m*e^=UP$uD|sp58#ckcG#4r*_zHar_LlTFCxktVby^d~zD
z@iWu7y(#a>rc8?8LFVbj5*@0ZUMVum-CgUi{>(KYl0%2>?~RJwLgsO}k9&wdh;6CV
z18LG2U{WuzNs)6xQepX&k*YnKBJ?W!kIEvQL@Tvn)Wiuy3)UVH-6fF{>4kf6<N=#R
z?+dq71HsmXhp?ZiQq8O0DXbX2P5j4&LvJ^%k5n+SxvutlE*|Lm1Lili0KXmm2VGP6
zop)$RiEfHEGe+n%b~W<1_R_e;lyX&zR=O%el)R_>RQ$$zv6PC|jx69Lp&6I05rv!B
zBz|v{;i_xX)W4!t!^;f=8--+olg^4FDVHm4%1z*Q84DumMrTP;%SWQ5QW)p%&eo(a
zN9U^(+0pFTFm1k;>MFx#|5Mj8f4J7e9D2VL>r6v`p@M8FY3MBHPk&ahs_-TQBj029
zrS^NOD9h9DGykybz<T{qghxHh<CKMk9XKyLJv<0m$(1WS3-a7m(YM@G@f4GeeFpAg
zfH}nS?1E4~*d|oqYQTDoP3|U9%t=^Ei?4=iJ4Zzax+g1@scMuG-h_{1;<&~fDah(W
zHi;GW0;85bQ*EGiEY8L!$;%^Y=-Bin$=$daOf{ny{X21wDWM*TWVA}|Cw>g@)02&7
z$h~k)>Ua1H=MMQqTeAb9`uN-MW8)WpWsZ(iD|E4Q!#i9bsY6COvkG<CXM`TpC2m=5
z1$q(n$(x=AVjELUV@w#=C*h;zI!1fXGSCU@wmS~4F*draY81=W^$-unx#nZDRSQ>(
z=mNHc2ZJ}-aH)wafojBlbM5fNI(KRVq_FZ5%pk@K6*bHd3n$~cXwn1v1c8iS3Ub`e
zg2VnSGL!11*Ne1cCXvaMi7Ny<^f#eTi5o(M=PY|gogVqq)t#y2&z44pdKepwE=(mQ
zS-l)Ml;1x1+B%%7rYr6eqiI2X<to;L98MSF_sC&JIbn6kOSG3u#K&B7eLnTxa+G^Y
zGzOL2%WZ7#A<tH0u6C0>ijJTrg?`{C`UfJjbdNSF=nTz8PC0b6iuR}Pvb>93t^SyB
zLRlfV5fiz_j4jgDeK$Oobdw$A_;5e=7MZGUU}g&)^3G#4=s|C^w&1*a6Z&l4HT4=f
ziE*pTqTO9)?v&mZ8yh_nIfC}I*H&H>ls8}1*BIS|82u@n$(}PZt6tJ`!Yv4oeuuhj
z>6N!Byofzay%SG~eIlpmhD>Yu3NqV%kDa8h-5n$RY?zqu!iyq(?N{yF@s-RDysdu)
z+EiPkRwnxLIdlv+BzPTs7jn~~(6`8XxL5h6J)^H7jfwHVW;o$E(yI3*F{zK}HMc~I
z&=!%3{*JWAPlX4Icd^M)D#S-)nV%hrrkd`pWf<eKevA%b%WNa@Yq<_Ao!$j*8XN2}
zM0<LHFo4J*UfRYpS)d|^@x6kX<c8cZ9I>|!*MJ_r_ne`LOU3?3hPp-W$K3IC)w&b6
zK@}l8Y^S!vi+U)ujP^PE;G1kwWRqT%wj19E&1?gB&{jod%aCq-=pZ~z_!ycSMQ`xj
z(|TZI%Mi((&<SLd@iypU;<Ow|=Y$BCDh38Hcl6<*smf~p(58aur}*RA3*Cow51qh{
z;j^9nTw}2S6p%|=4AK-<1=qMj^bWd0Pcd%62k@=V?oE|;pog0bx1Z!M*uDzW*+$e8
zJ|2_h`r4NOid+=?g?5IDLPyA5o+9R#a4tPmk4QZMO}<dQ?7rX^iFU?`UxZ1*Ds6%O
z1LLCmht^RZdXw>#-I!fB@`7GQnp{7I%v8hVE54ZQf$pnDPwb06)6qD5gYasDr1>Dj
zHHRn{X=IF+KiJ<Yjp%0~Ps(G!`z7ruUQ;hG4G4E)j&py|jC@q6NUfqTOJ4;?bTlgK
zpU52hcr76q$EV=mv$^!I%v7}pwH{QAjiOuVAEMhsRe>pdf_xFUBIP47Y#(m1eNdE$
zgmt&@gg~)*^h?37)n^ts9t3%Em3CFwMlJ-Mkh6Q4Jvn@yqqed{dqp(Dk0Wu~2C|FN
z34Ejd_%L*u_+D$It~S28FWNm?AGAGX$Y1zuWRoKxBqOyg7*QH{Dz%XY&?(?Ac?;h*
z)n)1o)4XGuh2oGf@8}u0ik?I#1{?D$z$k%;Y)7x>>1t<VZSHKxncVqUTlIzz!d7L?
zCxcXbB8)Wy8{>Rn4>~A}i;HZ{c)@;LYD1n>n>*{mjlvePG5MTGL<SK{9Tf8_n1TL}
zKO$9j|C2oqmSZJ9r8U+&sJ)C=J1^I7Za*M(BHIN*zAdD{PS=JLbaZ^$+I@iNjeHif
zl-0_8#iou?I`Loei&O<<ZW_UHa2B0`{zP5UBF@9D`+)7raBQ1mR&Gd5)EBU~F_?}G
zIjAgcer#j^1ABREh0skn0rnz%NG-=(Weai0ngN=~KZX0!=k!$tTPSItBk~_SkL&}z
z4Hf^-TT^{4^bQQi9-FHKw(Op~e<@y9Uu`s#)*;oT^WlczrkcaNQ94Oy!9T7_)GoO;
z)>6x8T#(XGD**BOa`F*4Ttkq%E0x%V!ImOos^^Q>FQL_~m4mC``EY%BFaI=pYDX1P
zB?bg*=u^d|{B`0G{F_`+(8zcM)*~NmKe4yq8Vod@FMJ9NhZW^#dKJuOqm+~K3FPgr
zY_gkBpnt|X!M0=_anQ~Q*b}rFS-}y*)099L9^Aseg5&i$b|Lv0RpK0)eAIgd?M}+f
zBeH48X?jGbrvAY6a=wF=)pNuheIq!C&H{VYuUL0yH+nig+t0{vz@^CIke^_%-)T%}
z06z$GnS}w5bU}>5ElP$`6v&Xg>{I|k(|7D=+J@W2EabC7S80S?s&{ZqBoEpc>Vn}F
z2NPbIBa6w~pq;QJcnGw}cVMOB3*az~AlLKTg;wZuwVDD$2iPTKydonl^?!LcItM&6
zEn(-$74S(U@9JZnB9{|=`uMW?)Pt1Bc=fdi%(&JO_6H8E3U(V#gdn>vb67boyh0f2
zh%%V0#up+>u%UPzaA8{q={Q~0kjXuGnwH%lM^VK%<DT6liTWIMhh+=Vm;3_Hl3(cM
zNKd(w9jJqWh&MMfU`g41U!5K%d?WiLF-?Cn8ipzgnN(JCns5yi7$t~HN&2NOUD0t*
z*-h`J>#!}9ZD5Bw2`qK|Y;LI$3U98DbXI>)=HNcMGT7%lf0!|M0^4lE&5H8H`4)_&
z3$-DROl&-Xs5xp4dY3RZ#{&C!+N`yir_1|nJB()Df5@xC_rzB6CzSDSPCJPsaEa0Z
z?Tt>@vxjMInn4&!Dg1$%N&aCQq8MSXa!5GC9I+ooO!`;vHFXePn$%sTEpf_4DIOc+
z_&_}(i`Bb!#@a8?jaa3hV=f}~#f3%N`L_pim^9E|J;n6(CK-$Ax!5^*rsYR-HF~~N
zpo_$FwjwYyb|vT}&*xD(Su0VV+xKI2Y|qeqWI}2QHD3D_JEx7RbXow$g<yZZk6vuG
z>&@_~Oj8V{Qu2RS&nQdi>x>>5N!*3An90H*qKftty_M<7*TqICkZwuY^yRQA+hR9G
zG*mpc=1B-{Qn}zojV2xXZo3<Hh1cv}gHR8WbsIkk4#TnRa+NMP9>XQ{X>klu9Y>I6
zB#P&NDa6m99qu4HfIs0cbQVEZ?8LVM3cMef+BCsPliU0IdsKX(m<AJ74$q`6Vk7ih
z0S8DAhLkOGb^JT?1F|o5%Q6i+gIq*blTGox?rn63;3vE#@t<QYbDoN$&bih^UO5H|
zcZjLV6!JWB4P}v7;g&y+R2-kB(e_NNXKLMjEr@F{#Ahcp!ynQ|l+}8h!WMLG<tFl#
zwCKlzG#H|v^TfJz^0d58Wt3LP1oCQO0@$R~E*Mda$kU|nLf61$?1+u`o6&Ay6gf$S
zj`2cGxXE$O7LR;3deA<3Gqw#4gDmVRx<=_G$P!dy{9~9WQV9vgME6(hFM7A)Rkq=l
zKsGi{sjNSdegn6OOa5u_YqPd;vJ@|_!ZtgPJ8!~zbS_z{B;iSFJFukSO2Nwmgnys>
z6-iY8qRZ)~njeG1mDi4`(irtBal-an@U!!)V8BtzBWftOOQD4nI9DIx#o&mRUrS2u
zbBSufULp%WVrP+)ureUw|4j4!hm>4oj{KWvz8vCzvplqSS8gO{$+BLIz6!S2o){P5
z9T<Zof!p{vaw@(>=zumSPs;ZbD^Tn7+AShN6-!ni>>0Le26@8QK^OSFu(vY^-lGkL
z?aGDPEq1iPUK{7N*nMk^TrilH=$86GDWl|=_7gHsE+xA-IvWYXYU2zx%W2m@r~_8b
zsH^W$9Td0sN3^mr7|XF<AUsH2{SQ(k{%qP=%+L$40WJ%_O3Xr*qLl45p3#`l|MCB)
z{So{Z&Zj0*hwyAzMSp?W6%W%e^2Sc)pCnGh`Hu2%n18K*wOyyS5HqCf;2oan`QH4N
z8A^5_7r_UnOi(Ot&{x|mfk5y?UR&E%x|P2-^*LaI2ej&Wb=4*G0c4_dEb$&hh+_Cq
zY(YQpT(*1zr?x%-Ph8^z<CMwp4&TR7jk~X&!fqpsF%x?&HYDEB&#`;dDypeA-jLE)
z+Zz?kXFCCjSd%al{~uJO+3E(bBrIT-Fl&@Upp6t~9EQ*B6zNtBOP#<3tUg;BdX065
zvw=ds(6*wKoM`W*ISo?Xe|V;DfI%c{uD16B^S%Eq%qT+_QSqheNGOxe4EI4-XdALm
zU|$?BusOseZ+>a9s}9WyOUPBQw!^Y7qCJpCLWZ47E)0$UBYi!vd5Dc3VzDYH{5|xy
z?J#*$8<93stz|oFd`~##l!7S<Nn&l=NTiB<QKqycVrNbTq9rnVFGXyE<%x&<TYNgB
z(JKnZX}!you~?}~(p?&oPIyPL>+$#miN+tI`z+;>jc^5Ydf;q$pU-6)%`Vo5;9jw5
z#H4oTaBvt#g-1$H%L(oGKt<#!))~j~rebelT*Tv;u7vbb;U!!R41O?j49v5h3S7)C
z5S++cX)I_5*AR9l2Axi3Fs&%F{*TOo*>qPS2Y!L|^4f?;0|I?4j?q^qm(?tBv;H)t
zyAp;5b=y(iSy{<|J&9D^YuwE49_WL_;u9Pu?JjvrcR-)9O&<vwX_sM!eP7-SmSJ{d
zFR)4uD9#JW;uuh{<(NNO#){tvzDQbc9nTTuB;#4*8bnVx)!$64WlO`3>4$u8RgLUe
z)|37PyMc<rHdvSbN<C#*>vXygSwV@3I*FHj4|49#H_)x*;7%o5x+4(6hNNS`x<YfN
zs&vBkfImwN@G43z<&tp~4i+1g<v|^JGg_#a>FH{9ex^T!bcR)gCvcin3S#gi@<$^9
zAL2+CzIyTWn`$||pV%d#Ho02;%X<^)fcNr`1jYO&YY&@^u7~5!QTaZ734O&8M7oxJ
zS+D2%Cn|G7AJoNqp|&)xjpZG=KBp;OO^7;Qp&9iiCzKO^6aJM<XmcupOhX1xAKCRm
zUFj?x%F0r1+5eFD;cJ}Z;f%cXLNeM5Sxom6qvUbaVe9XCg|AmHi?x&r^mu8Sd{R4s
zrb_p6I!gnj14stYwT7Sx?2kXwzkyBIjpTUR4aPZ^p*3KZ&{3Ic=E>fne@!#A9?Gy_
zCu~xrC!S4vy>FCj=y~-pQ%Svs?^1X04PvLTAEdp+vz+NvOHV^RRr(s-PyVZKcN8ig
z>`SEdNG*Cbb^`k&cn`gM@U{I5c<w9*2HG0w*~)8TL+%UYi1!?EP#8kv^hhEO-VD7V
zN3=|WJB1?Ql6k3w`1+U%#H-3z>m%D>dZR+CnXs;9BK`|u#>X40q#?q;RIE-(xLP4J
zOuokaK(3Gr=@$B7w2MCfrzAc$;c#Gd)A9<b)fY3DS77Jc7pl2H!yeMWgvw=wL%J-d
z(0%QL?QM{Ko{mODIjZc#Pmx3Eo1wR!-M*ic{d;YJsjj1zdahpLJbWse36jzu5Hp>V
zK&tbyk&#f{SrjmA_teg&hHzE%B5b164m}ASL&m5h<=>>yxx3*Zcm)}ZJ))G|erG4T
zu5C0;Ql0SMyq)nx`a$j_h|5QT+kajxGMagsgC^>2^F=lu^g~*+n1sr+lx0Xyypg;+
zFfno^zPtU5`jL1^*ChqUkw=m%;R1w5{-IZcNmvHoB3L!?k#mtU+)<z&X4feT(M$R*
zI15Rp1F;FzEjWW7kai>3HL^7^2FY^9pcBw`%FSqHey-}H=g?ORl3<`rwXf_tX&OV)
zwn*|ZUlv(kyc>9KNEB+z2MM8vVt%9+aR6Torzani-vpO68%?Kyrim^hOKOF4q^$Nt
zI(shEZOF8+E>KONHM5HT5znFCkz1(@BtYyyU9Hx!3)Mz+ccDj6v7ZJ1GiGTAZJj}h
zK9wH1_mWU6pn;mvyud3do;^=aRtw=MsWlia#70W_&PFi;Kud6sJ{_!|@Ph1OvB+Qf
zx!|moQE<H@H5QvswxOpdZlO|amGtd$LuIZoRXCGeo%lhB!p>A1J<XOW_5&S}bK#}R
zPuhXhXYvK)I?_s?W*<NWlwXL}a8mN&rgs349>IM?A1OdqCJWd<sIlzUw7JA?b+q^r
z4AuK-bKndGrz-N#!E7I)ZKiIKeX&3D`|`JmU*J#Let2X5659ei6Y+R%(ASC8jTfR>
z@CkcPiC6j%8$kKcFfi13fmAUsRl4in)8j=)QBeI`93}2x03?GuVPmqR+Gg`W=k+vJ
z%*QFJx#y`+i{1sFd$U|0wSIveWj*-Wq;`SoXba0L+bNnB(>#BxsX!HvAs(V(WEb@+
zFUR2^J_={FIaF_Imj73L8QEGbq!d#VVgfTwZ=jE(jOhK4DrV=$SV&(3%HdCi^Bd2A
z-Ng-LhuwB{JpRS;t7HrO&DN#HqQ|tO-jU(A*ahSOOrU3@ClEhqsP{uY!(n<xS-!Hy
z5dH!)P``_;*7C$o;%%`Qevbu|cBvDUKIUeI54O_Ex;g)4R}?m5n@Z=AFLWa8tXPDr
zHZ$EiXaTR~F5m`Lgyito@+<7P;2$V&3T%P%1B<N#<;~J3M+EBxDw^jB8T_t-`LH`N
z)w0}+(eE7toZknB01YWX&xq}We(<*0BECQ;qFvxz=-@MvxZNY+RYX!vd?h?h>Xfv2
zhfO*`EteA-H=(LoPe^~lYw|{U4_a9|9Gq*q?b+u09-oR_5;FI^4ouaSnOiH%a*x?7
z!K6+7v_TC}8)a?@e^)mLlE|>gnTnBS=mR)MJT1}WQ**i1vlJ6D#Pbw>a;`Le6BPLw
zWUwlZxt3YTSU;cBS}ZXAE|`7IJwEfkf)Z(x%|%!PXH2(5r|>cFp<0YIEGP_)b~r&Z
zsRVBahofJ~JhCG3u@R9UBUQxX#7V>}=RmNb>=vmixB|ENt>O(NIWSYWqet>GgBQg?
zo(g&gw4suRq!qjsH~2MevA8fHkhd>C8zeN^Xa7@d$WQfGOdgU{2>&3un?70og9C^~
zcz)h7?6z~0DY<FVu7uq)VTPv$ykqn6U6N*4=LhzYHs4vbIo3m|;_Rr*0YioN9)&!T
zSWzo<j`8;tPJ>pv*I*&jJL^ZWnXgD*9hA#B$HvLi3bvH>cJG3d9m!=ZWk2PO4ahe=
zC(zdTV6nN9MgAxgLNA-c?sRlOrkL7+rp8&PMVP*=JJJX23e>eTa;bls?P<0fZECF#
zSKB%g7RoH;iHXP-UlsVu-;k&!-wDq5jYJ0Hx0N2+`hpJP_ofW<Mt|kt4Ssr@WPXhd
z!s;OP%-7LFaI_lBe=fUoPC6*zSJa8-!M3(?`h?(XkH!BCJ1%uZI*{#=nf&4E|H8+%
z53wf4xTJvPXCY5a0gTfxtt7M5$ANN|7;A^!G@2v)sALvQKrZa=2GG1CV3rgK+Dx^?
zQsI7Jd9V?jWu>JqGAx*a{D4)G7AwM@n#n`OZ%OHVH~;_St-v4X0^1c-Mt<;@+qy0A
zF@CJSx%DSy53GX4%K1V(8X#BLUm?4Ks8{&EKS`{_OK>)E&RSjED)#34+Y^J|iK~2#
zv{8ZPULu&Xr;5}|$rN8|n|A$Xs|fB}62%1R8CE&)E5DRK0EY)k`2k>gKo@(<-^3H%
zk?=V7O$cKZY~AceaFBDt&gw)pV=VETP(wKvJSX?r`?F^ol8;^%?g&e?HNkblDl?zd
z&iBezSsK0fvNXWd!-vXcN~BQTH-fAan<iWr=ZGt`eE0+(CM^Y<e5~q`+E{v+2O<j!
z?mKo9IS$@QI^V>P2~whD;hlNnWAP@QE*!+B1q@6CJ>{i#%6H%L)|V$w_s_uAVPbaJ
z5(P8VH+;v0dBQECKFx~;_>Sk|zJM;6DL-xdU2g?(5phlVJ6KV@lF}2M8hi*(1wMgx
z$^va0zsmZTRMFg+eD5ceC_3J2N@)UHINB1SzyT-7ui;&<nY=nuB95X4#e9Pq_;n;V
zr=P97FfggQa2hTo!~9o!C3{-_UHVlp-@(B8!3)GdUh(nRX?Py&7rew}2a2`GZ)rOo
z><6aVtHEK`pYYs*+DMW!&i|9+3TSNW<aQEy#8na#wm6~=%z8ZEg?#`)G0ab+tn@r(
zGxkw@rL~hK26Hb~{v_wcwgdx|Hr8bY1qm;3hpnO<ggu?z!j)(pv`Mfbhm&skEz*+K
z0}v~6H?68CLcDF*uHbjecz$1e4-SzpC>hu;JydWPPHAk%imWZ2&xrOg8~u@}Zpy;m
zBn~uwLNoEYNzI+_QIAkB$XS=tF<^|)6FCqZfVOp95K9u8d&e__Q?9~$&K&1DNJwV#
z0ay%Lfwo`?tnb|C$yE9vosjGCDZ*a3mrcSF)e=V=^fuN{yF+e<g4)VzR|krpY{T(~
zi6vy5K1E&@XNbl4CcT-^%6he+h4ioOvOOeTwts-tkzedbgcr_Di7tuLdl5%re_N|Y
zE^i#N2z5D^+HNRb`$>4y(@`x!XNisJ0^1<x3(sq1KK}(SlHN9gAQj~MS7W&cN-f9i
z*5ry_rxN2$Ytk6(AqFEAv=rM4*=LExLP?|3KIU!mwzn<FpR649bVXB{ih(rst~f|R
zv2R2w*eY#RkI^sE70_6kEp703&09tc1a_pWeW~mQDEM7&=4k^e;s0UVgLd2niu}vK
zG4zhS&UZ)}=dc9NlCw3hy)ik=G#|#;5j{STCRt5ObGMQSXo*wy$Ehm`1=a_LrEGkG
zI30F!c!E2Gp5O@r8BxsjMI6(Fp<pVW0RUOI@Z%<xdhS?AZ9?1XOP!<C#e6sA7Bar+
z2AHR(J8|1e?7ZVIX^eHT)||EEn#l@fGj3CnVZs_y;hvd<Po3?kf*f{Kb5zu3B|aef
zmYvTd`JDfm)MmHK))864SC*IOtHdhD3Pi|yEH$@ep)qRJ{0!s{vH->yiDmSyjW`LD
z=n(RZIF0p0nt5+w7n0ub3muMqe<$C`SxTNSJZ66q`_0<g`5ZHp`}VJnxl}BM+PX@M
z;AmopwViLh<wHO@bPHY*7NbM3W5G$p6)YxD9r?p~9;x;-qE-gC)&HD4ZUINguJLP-
z93>5mR$`H(<QVL&ZKCb2Z3+6>UC}?4O3ZnxW>af}XYqK&PdCp`G9NAY72d?=V#CP$
z_)XI^rk!#CeG2x#{!$E?Ck%&gkW5Q=<ce~#Ov&Kj;`rFufu48H1!S?r!wU#z?QX<5
z^#kRc&Cq?;c64X)A4rQm<tFk~aV<GRZtOpmbV6zE`y8}M8{od&2K-G=KmR^qg0s4E
z+&L7vU0{iRAWH4a3z~^}dN(2yWbauL)EwoMCA`JxSMUd&#m7mLpbIwhKgMbrFgFfw
z0BvA~Qi>1Oo+WNp&Lq6>)KUKqj;8EFbI%Se8DB|MBYM~GPVSaI2A*wMhHb;GVm&ZG
z+5t1*G>63!^zOwiehRCgSo}WwYH5aTFnDi$3VY$#aKtuwZ@y!j-W4>{uAB1`UOIU#
z8CJ4xmTvE@v}1W$?pKpq2z%vo3B5qu;49}N{;T&mH4kPAwUuO|1nY!NftyS*XjSkh
zvd3;##?z~abJ{cKXDJ6SZdwT(^A-R$aKt$Uw7`ma&aqUS%5M*<ehgUzT46=Pwxk|j
z198#iK)FJCM<ZaDz}}Box>DWYiPVNA#4q6wY5{sss!it;5v;D0BA+|f!>jy$v1?wA
zO0fS0omwgVAev_R4iu4ISOb*99y|L{Rro=EfQ*5?{GGLC_+qU<obBo6-{)^+$%vaS
z{o$-n?3Bl&GmYVviOywG88$={rMt!Zh$vQFRz3LzGvPq=v2q%w+lunelZ?^=N!5$Q
zYUJ!4mvCFKMZyg6T<B!5XZ`K+V@lH9>P7QE2TSBy#0_DzIu-i{o0FWhMHz(jrBB<t
z(B&LgJZA#OVX)wUt!AJDKNjsNeBvG0Y9tZ4z!^nbK^H<br{JqR1*l3c)RG*-q#u>?
zGMz8s^WmV_^Y&}loA@%#LOf#1SBvp40SfJ5KZmtPs@Zzw;P8-xG3};e<t6G0<cXA|
zM-$HAajr<&<kEQQj=GH>ratnoriz3Li8`N|^tUz+Gmy9TEB>bT?!FJmrs#U;LNC}a
zi!@(Bs&WVscOnDOdHTYFC!ROn`@rTfAx?7TzBXz^tozH~b<%AXw6T&e{%7~`&+VCj
zkUsGXa`WN4#Eo#P&xai>nD01<uEK`-&&n@7oxxopRT;VRy>b@YD{KqAf|Y#jB(FbT
zzU@Eh*_4+fB@^TQz%j{J)zy$3s%$ZP<&(Cy*n6dS;u&qDo~f5Z%gL3n$><BICSKb&
zQK%vIN*pcR5;<`r5mTng?8K&MS8}uYMe4J_dz|J8zZI(muh=_S%*synvVT82fvy_o
z@h2gdEUn`1*sp`-WOr|tuoiEQiSh<wY~Yjgim;7O*|%DJCuKK1;moKr2YZM7pd1st
z)>mZ+Z@iEx^*|;os}dUKR&ElI2Ah&WGh{na8%tNK5fif80@bq+9<sj%V*_!{)?%_&
zLCQU_A&`VFcg!Ky5exC1;!g2;!ArlHOmf!3`Z^yv{)be{&k=Sx?o)p02Vp^C%>$kE
ze6wFW;iw>Hk)6qV$YQdFICJkR{5!Rr2rLW9ANVt%du(OprelwDm!Oha<ge6I!EO&F
z&xAkF0ohRY*d1!BGa#(OzDh#YG%2wADnATf0QLQgwBD*m9Ae*5uoH=w>)Lg)Lg+Jo
zLih$&)E|H~Cs?G|xij$y(gj(kAZWJjS9GxPU9g_LSd7C;&{ud1BEk7F*odg>nB*N+
zD8sjj4UtyrYP3eeT*qONC8~=KK3=R4yh_(3>ytLmHL#s%;VYy|?Yi*AUl~7YYmNxu
z6qev?LC4S`Ac6pqjAX|zCsDd7s3Ek+7MQ0?qt%&Zr`(;lVBh<G7Rq9sFDKk4&nhF;
zBsrE>or^sEY(L@Ekrv{9b&q8Ydf%2q4Ti6FEbzO@Ax%SIqzr(~+(pf?=Q|JBrh7y7
zx8fEsC@DLz7k2U=<4NTj{x@<tsf~RkRt2vK$7mk>yl}~XP57cu2__?-v{4Dm;0-h$
zf#$?u(5PbFf%veB=GD<b!FT8m@or)};SsI*${1#R@!rqU^uT#?u~-vy(>ub^fN@}0
z{@%j`Cg0fH986TAM5p+(&X;A=|C>z~=S+xMY{YY-8`fWWr}su#I?eeN<CGmcYw|Z*
zS0dMl$>ar5q^_bP^kHRnaV&9Db`q5IfGD=s16D_QxsK%E9saRpDCHeKi6{t8*m-?V
z4^LZchb@Fx4UCeq9Cj^;<jE`G5$FE`AGKOyW78Ue=M#bnU~+stFxlS+>+Ia49(IP6
z-dK;oL31{G8@x&sKo+Q~(6L3<0azWZ^PYxrB94Qg!&TO;m8<d-sDkxHe#IvWg)q(D
zO|B%Csp#+?e8i4L=VA#&f8|bsy2}FI(*68f0Ooa*D+BA5gPuWw4NWI#qH)Gvk`Mxw
z{ReH&trkx%zk&KbyN&!0-+|f%QJCY8({G7gu&v^wL_vHGRQ`2ARmxvbPML_MIP+}J
zY#rloupNVM<yUzc5#`e{npm#&lX^9>(o={A=sS5c+%E>Dd|Ma%RPz*lS6PR9F!!rB
zmZm*Ztv85Me%%t01xvB`87&t*C*MPB+U`-^^mG1&N>_E5|D$si$xu1GEAi*DE^ySw
zsq!w~nRrt8X!+i|9vp{vbHCH-!aWEZNZa`j#@VhEOhpXGATSU2Aiez;taBWfQ8aNg
zPm5>$hjNk<x|qi{%&;$XRB+sLR8P!w+(Wk+AH+ZNZth)SzfwBKJOq6K=JKgw6pY>S
zSXmQnt?s}VN?#+D%GAp$R$ea;UdS6z6h*)%%W{7eag)EYWe3t;#Y6*|z$e>x@k;_r
z!8Bwx@)N(sV+H~7J5mT;k!qW#CoCXW@oCnOw3S##_9I(bz5H76pUn+57`r!;(9V1Y
z8A9}Ga62?ksb%vA4}$%8ifbJ91se>8IO>=$$r%ERoyxDnAoyM678--MslO7Q#n$C!
z)S3rZ=zW~&!6|%sN1_DcJ$O#ANhHPj-hWUSYF&-JW*_@+OW*0u@llct<KPB!rtO<J
z8e+PRPau=>8-oJJQ7ktwI(~^G22RCJpx@L!iTA)^Gq!h#_8wbb@KxTHyA7{JX!1y{
zsd>J=iSj;qg7J{AhBSlUDYeQp@tNHFh@x+R7W^NjpS<2*RB%YWOg{0f#s3oaNIkTB
z;y1WB_|Z}iiCBB9y*B;<P4xyU#XD{ycgU4!AIVSq$u!GwFk7mPB$OdjL;056!S+GG
z3QXk}J6nQ6d>~zf=fXz*B_Pe_Cx%dch<aoO8DoDUZ6a4Wmip+8b%;uGV@uV%T5!60
z6nQBSdPn;y=Y=hYVTpW)^!Y#UewZ^rZq43ldQLIhPhnY3v-aRXP&C6gehk|Tcx%l)
z0pvzrEptoatdz04r^VtZ{kbnO1|KDa{Ws0U#Lgm`_|4}-zw*C=kTRB*%(Vl5QQv_n
zLYq3AbpkQIEC=QbsxUQ)r@D)E#0tb_{t5cfdeKr2Y_->Jl1JE}hRqFh;&hEiGfAZ^
z#ydD`sjV1BYq;}`;}7zqnu$=RL~&;s)3m~?1SaJ6G3|gRa!a{uNK0yt*2eo^@ELJR
zZ0)U~{SUkrO$AkLx|4#-kt8@veB<dTnh{c^t!MBb>2s!@$|US<g9vIBo#Zg*Nbx0c
zOK+!?`eyN0;X;oD1@jhRk$AZb`7Od9677(t{2R^TkgRis-cF}*oEW&bG9Ll2_u~E<
zV1l<Y@g6P&yX9EZ>bOg3fr7(IWi8H{CYH~CW$)+sDxdT=Q@eXc^Zz-w;cjUe3X!MI
zba_+mL)+QB+v4A*%uESukNzGPgUk)^Nxh_q=`(glYy%+ZV{fYu7uqz5;XnJ<IgbRp
z2<dX0f*N>?uUEa?GDdBX)Bq{2e>%UCG(%{GZ$UtCz3-AO6MyfEc)P(zj>6c8z@bAD
zuA$EbG|yqZ>*(Rp3+mdYp|W@^wx`XEKHFP6DH%OnWgI^l9xS+SeUoUjQ{q~I6`ltA
z?O2$5)W6>0Mcc$}i9L>f^8I3dB(3pZ@F(m@5~q0X<pzXFuvt(B*Gzo_`L@dDWN{Kd
z!FolQr#uWaPI@mqFH^Fo`RCxFV4Qf*)LpPS%9(S#tAq^=HvkiI)L+FC5VxZ19UJ0r
zc+Ut&?B{o##=nzlqK&P;lcOE|VG#bFxVntJ!jx~e0&$X~Z$f)Cjy!<oh~1Ha(ol4a
z<Dl7+PzCWgKKq^WRX94&jB@px?H%k*5cY|K1v3Aw>}py8j7LlXTx^nb14*;*Ko;jN
zN9w{wLIQud#vITKLz-j;D>>>1uEkcCZsu>1w+Y2~G!Ww%XgZE{68q<->;~|WsQB;O
zF3G2zl@tt&55DlVcd+27JVzKq9X4lU1H|jX2onin<Tb)4{8;R!V0M0U{uk4GAA#MI
zYRYx-bCwZU3jEjFMfhsDV{h)%`3qTAp^O*s)lcpQx_bM29-zO%v1JNiedUPhOU|X>
zn8X>P#naheUz#gs3f&S3_}sAo-fIl<<=CBI?Y7gl2cFsfo`Lqz7Yy0P__pC^O$!4$
zKNtBiZyjux)Yv!Myf3y6x(FYHFv#%y?t%09AEfH|b^D;*uHX#eabTe-&1u+XCnq}x
zC3Py8?`#Xw{f5v*+9m#Ad*T_np}KS;&W3s|tvwSgM{TNQs584@mi)x?Gj9ua4eqxd
z7i~5NCJ3dT|JjZUQGZ0L<ZFaY_ms+u1M?&p_!&tS=Zca?us;fP5C++wS=#ucVpVLO
ze~*Kcc>D8$yyOmh-UJs1y4d1OB;G=KDr}I?;FIlr<vhHa*d2@T^#gq3`haF_x963-
zrB)-wV$G8a_)QXJ!E#4fZ_6`-OD&mAs#%%=i#Ht^6Z{3)E>Cl2!Mz0u&KAx;lV+gF
zf>(ShH@6&dPV(paXQJ=K^HQ&R>0qtDhNWy5wL4!0juGxTdkKkXcHRlwWWM~~EyQ|r
z9qDY+bxBDc*)+~F#{3?gAS@Ke1mZp0AP^V$vVF~kkg!U$tsRyeMvnMi<(<W|_%`|P
z1Cn!m{#$eR+-|lgoFnkwim-yz&r{X3&xyFo{Dl#ZYm2x#+8&u?)C*-%eImV-PT}oD
zB>8*lRGrrPvcegW=8UF>4IR`+S4R<DXxaryaN5|<a<LSvECBcPc;zGcC}n0E2mjF9
zbL}HP(+kSD)rrBsUHOsLky5=HbRYrUS=2(euzv_+Qv1;BLv4(%g}tckf;Aj0b15xL
zP1bD|FPB}EMcVu^+IzY_x-PAUz6drEE;J80BH2Q53V$%VIC4r~6FK1)(#{lC3UbUQ
zVRO_GeZ*ZYn~-zsJ3?8hul*hIZXm+@z{T(etTD{j)iOuN)ab9=DRzsg9}}XYbX%gb
z-h>^bKMW+WpHp*`mqjD66QCXY7CL#xC}oEiT~A4Kw{uk?UhSIaW|`S3ySU4?WgG?S
z7#+Cu)PQqB%7~O{e3)&`O)L6UyX2bY!c+dF1F5Hl7gU-5k)CCYO_}W)5RPi|Xx!n8
zwq)(*E0MM6{>V{gyytjYlW^P8-bxGj9aYhDGu+fYS+9%nY{h_^dXAHkSJ>%&ebRQM
z)rkC@63_IsO-wH;9`BS2Po@lwRAS0=?eaGewc+Q~B67Ryh5_lGsfc?#-Goc!9!578
zzT_@ZMQ|$@qy9ye;|yjuy*8Q(Ye%Yv3d$Urcj#u3IfYA8bGcWc{@OLY3sJ%Sv(Kz|
zkoJWpp}A$V{WshVim#y4+yWh6^uc*b?ykRQvjd~l3(@@4zr1$(ATz1VN$g@zg?F-(
z*$PN+<uBG%_ynZTC*cHiZTPKnR_&%AEQ~jwCKKpV`iLtSNZ}@^5KK!Ef@k5)l=pJf
zn2d9ng6t1pAzzR&(Lh=sql&j{q+B?b+Cbe8O(|{_{tXGjRcUS9Q$XeDOL;R>=I`XF
z^i|qIeW>vux;AY^nRj+)nL%eH^^@B|J=U5b%iJM;WY9{UKo<$IR0VcX(Fm@EKF@f-
zeGD)~7A`$)X5nb}>}cI03%P(v)|*GCi=}Gc==V~kq<P_EMUPS+rED*oA0Z3RrOaS|
zDUJ&j>C@4Y=$`1{;(a00`nj|#^3K&H+9Mik3=DP6PT)AUOX0YXAFl*`I!wmF3+y%L
zkWL2E;23wvjYLE`QT-+ISD>%_Z{$zLLhf>e9Lu<~p-QUX9d>1wUz)N_n6FPJF8le=
zR=siK#-WTzeyAksaev}6!6R4Y@O+_fI8$pE-l7d6|FOEXWu;vM@!I>Ss?4AlhqByb
z^@9MawWALA<AU+g2OwEHXe=(erED$p|7FwLN}mOOCi>wk$s5XlNIum*)Rw!3ykjx<
z3FC4>6*@2SNG1y}!Wtz#-7kzOuzFmC9vS^bY)dTFUx;nFSuK80NU{~ZQVGHZ4l;Tp
zLa2wuSl26sNQptWXcxn~m=xwVvp2GqZN#L6`cl6FXC&?aadejPO`U%mhvKY7N-1@X
z9Jib##l5)0kO9M83d4tV42MCnQlOHgk>i$=w3OoRI$*$X8#dg=*mM5R`@G2K&bd#1
za$Voco~I}1bp>BqnRG4x5<M}TWs6hikk#~SYACA<xG|+g(|fiNWgSX#L(HZL1z)tP
z%rRO~=9ZzJ1big=0fC6a;b)+dwFXa8H#_3(SLulwPA98X^d|Te^BB*>%4u|u@+i8q
za}+s^TrafQun7Xu3aM7OUitotli|wJCq%3dk@llEOds?|^k}jNwU(YuMA<6%8yPm@
zY)P_GSDzS~7wRte487X*hPzX|fgFUNhUSYC!OM7^(4XNZQh=-}m(oG{IM7vDz<cd?
zg{*KKIy#uG|7XH%9l$}<Bz-G2P+n=ft;Us2LaNIR!cF*2VjE?my}N@@UWFG(FFh|U
zmu)^O0qL*rly34D?7f*RXQWz4Y{u*RFl}*YGw~_-$@`2RQC`LLJgXsooi26dQDRvy
zVpi6=;3aa0o=%T3JlErc6-t++6sb*Mv06^M>~lh+gXy-5!YA89+9Y(#YN20)UI!Zm
zY~dnleO5i88@VdnNZ84|Q_kAg2FLL&oeS;N%GT(EwO;xe>ZI_eEl(Ik#tWmNk(3?X
z0XJ{f;5Rc`nH8jd(m_X7@#I|-@fM*|)Nj%-`6F>OP^Mh67trT$Us<8;&6W$nELHGz
z5<z=ys+%^4O+acWH<UWT3uXNY#%D}Mgi%aYx)Zx!ziK}M?8g=e?ew4JX}ey8ddQdT
znkh~DS3NBCY*t_E8tkbrv;7$Ct4fBOeltEuCy3R`Z}^nZUDhV7wd+8!v%2Gta9cV>
zn5$mFE-E!Nwu~;HjhmS2^a}hmf5mxTJ?A{-I7Wlm4)D9saq2XA#Zc=dI;W${u+p+<
zs*7}1GL=n-yJMf^St-krD6uN>t1Vy4*IJ6-IR?Y0@iDZyY%|eOJr<lmwSrC&l-N;g
z5>`U*@O?zhtm<@Q=N!Aox;{9{|29<18WoDPuNJytmoTlso^>PGMc>u*xz;^+%{(02
zLq%9$?)H=QMASD{?}0qCYvN7$T<8^BsVr(5uQVd_wO;foduJ7wt|~qGhW2a;;I7k2
zfo!vrJ{s6V7uzDUWMFadZP{_sP2Hj9I{prIp*w|=fg92|%OCI@B}43JZ=#(oZ>BeB
z(p)MqPCQaPYl#DHs1KxumOr&5Y=^Qc{)%2oyOsB$#`HS1wtakgZ79cn-cdhPASTHj
z>?4uy<FYV5yq#X--IzsEuB`S1hQx_?;0Sd#Ji>6}Cx&*V7pdF%USX#$@e7?x?X86F
za;0g5-cra6CnAf`uh6Bk_S$LtL;Ak(nAC$4#hEQn(yio$+9@hh?MGh9DkQh-2V;Ed
z34J$_6CUARg#4ZahwxDS-PwlDtsHCP7$~(3FV*k6WXC#MQV3)!wVlp`Q<Vnv7%GQ6
zT#$)m*+S9{Ni4hK=xb|(y)CoLd$?FSOE*(_QpfOgM<MYjyje|1ZV+z6b}MIWSA_H2
zNVrirW|ui!(AfBA)CK2ztvFDO=moFAuFf8My6}+Qj#12YsjW;a*MmzXTR4(?10T^Z
zX-C=7@EpQHmZmRue57-2oG^eGp!e2)q_>tU+Q(q?;5GZtdSX@#zodCvC6c8pju;x-
z&Y{iW`AWCYXn!;MR}xnn*ju2mWpcnSY<Dh|?f_4uf0g$3k$NSa<h+2-3|zMDw9P5G
zq}QZA;S%x4Ptg5!M{;8<C;D*G5q+$GjXuQwlMrp!nfP!UC<4OdzlEv%Aap#x#^azv
z>NBDpy<Tgp{4Q@s`jYX;QtQL=wn_^&Ko8*uIQL^)>2K6tL#Z6l-6*2A3D&j2#3u72
z$2C1~wSw1T7Si{Cb@~@=B6dP~q0Dve4(`!f(!0%D<o(1ceXyQsIP2D^c5r&^CncBs
zsK($!^&KHY4{CoEoEREJ+R*l8^Gau`b)W~nMWGx5c8`@05x?o()9>5Lxo={7{;l3n
zG8%I9nZ2i7wD)#Y3%}&L+1q5~Yc9PJK8(!sPd8D{zGUCBb6TD>TWLi{(<_}n1XF^R
ztddZi*4_D*|G-k<6*Wn&L;S7`q`}aQtV_DlBdIgM2JCFz5nK{{$i`B$LRB0+919$^
z!+H9avf=g<Xl-L|m_nSUl@#O5oqu5eVO7d+D7mH00%83ao<jacCMpMs_V%yRd1(ms
z&b|%$O*&$qV)q$aoMuQQ=p{PBn`MVmov3HW?N;#tVOIyEU%20d0$Ve^IuR>&mg`fK
zl#}+R*r#B$D_7l8`c|tNTIT*)?@9F5vg~Wk>zs>9+XU@+2H6aaw7Ih`()I9e!HQrd
zo)MZxR4>l5L&!z$tiLUN%h$s-E4;~4f`Q@AIFD6OUU83-khWbZ1h0~t1m3aFQ9uKz
zDbz@KOytDAMpoII(+h0t%WlF=B$rm^-xf@jA5jYPOx>;Ck<3X68Ae4%FQhx$PKm41
z#+eV;$L<2X^ROh;t85{+3pem#_QBd^Ye+xs=`6tVKTKh$dDi+CwU{c-Z`x4%8FEY6
z2z^~Rj;e}G3Ul_3nN#(1xZqr^HVu{qBj{5yT+ZkP&>M1#c{u1mE~$ISXUb3XKKe>%
zlA)yD8D0afq8WCy+M8@8A1#OsL*co4^#oTes2x^vsbfk~&?|o+YZG4Njtf*9gj+j?
zl6O#CPAxqz@8xRRAIeU8u)fluj`X(p&=T83wT&|`n5%pfUXgWe@wO9Ke`QUmADv&;
zlkaGKZd<NjrZH}$(+vD5^bK}W=9KnRdIJ}f%S2cBig|QaP=CIC5k01OoNXsn&(@S0
zt39W}N}4H16;LZ3BdNl&W$31?+i<IpRT0%k_K(gg$T0iH3|;zO8!GhmcSep<m&?Oi
zz7lVn5!AQWZ{8E0uE1ebpFwQF@|3sIT4k2JAyk{3s`pd6(>dZ@gSGMt5<xjb{p{_K
zSi~pg>|(Q0$o5iWZnZQ*o2zUIhxGlS=F+ON1bU-Zq#O;UYbXxdQ<PJ>$=Ozo2Wo&d
z?K6XY=tMiFE9ptW=1xGVM?Qh>+SkiZgC6=A`;zDgRVcaTb<0O;u>LF7-{cNWv)R-*
z<gK+!d9&iq5>uK)d`CUEUo`!MIh^khglJ@+08J8N>}M%mPgldv>Qo!Y3c^`FUL9$R
zbYkWG$Ta5)TN3;guBvX9+wg0sfZ?>f#6KqHQoqnoLyJ;q`}@p8#4zQcbC@Nv?5_Mm
z7|SM*kJV;mU83CnpLE>WQ%S^E+Fz(_=&;s?*12BvGheH+HTLsjW81^fpV&2GxcQ6K
zPww)qq1)iYun9`JQlC_%-)&36nQ0@!JIy`9>(Rqv2m3$NEMb(sxxBV5f}R}wmHte0
z6Dq<P`V2A0;J7>{DY>aS+;GPow90gJS{@pvmOCT$tHd$yG3kFQ&U17!IfKv68csVA
z7rQd_6rV07>Q9mVaEd&Yx?;aVpW4MZpTiM)7k!xXDOwP^N@lCagWu~LmC=sc&j0j{
zwqL^o^&I+T*&$nZ-KrhXcaR71e}r+?pTtoDN9`%wrq9xHgdF0Kenv}l-P<)m9dG-^
zwp4ng|6w?vK>LsMG~#pUD6&&|N-R?1$T)9vuy5s1a+TB^P?2JMjC9P_RPOCakuK}!
z99FU)R%zcJ{9Eg0?<bd#O+hO+r?4ZjpB`*)Y<q=5dVubx?atbbZPx}nl9>kKp23;L
zAHze5f3S{aoz<Im!IE3v-4t6+ws28C+k)`F^p2okU!?WX6N0*VufC30%oiF@*aqod
zT7q~ENg@~1yAg+Qjyw$h9opco4S32%C@>}~otdAghr#zmUdr%bbGdrBrMe#KKzB}g
zPP|3h0cNaUs7PsI+9Xzi(}?dtOno6%p<hXzp%>~-suwYZJS5cez9oNf7>Z!p;`=KE
za68q!P@ys_rismJyr*{wb<z<lpV=5Ygtb?*Q(H*yEn`UwHJQ7?MucWED$>(8PoWW-
zYmQx^w-^+L_C!$oLR_UUvuA*S*gdoYgmye)MQ(`eRXh}A<ZX^Y$Qq51O8tsu8Ni86
zloj5VwjqYE<uenhKL=I`+1d<bskz0@b5wUOhnXRqqb_*ISjwrZR1$#W0VP)ciT?}S
zf&T(kroFS@G8|@C1iRiYgsTIH_HgIYImAGrnY}?Umu_HqCq63MY=h;+bfox;yd@n_
zo^V6O&-R1#Y;~2kBm4)&TW+Sl)fB3!vfS_qu1fDM4^vu5kCpX~ZX(ZjBkBr0=v6>>
zqhr0K#@fGPJu<JLoO3hUlJ4eD(jvuTuoZ@@uYhypd3}Q3JET%I>|L3M(sSUEhH;6G
z75I3=dl1$p=_jNs($Ug)!E=nn_Tp-aeT6ytzTh&|B_^m#gOR$A>Z+#7Emd4Njb~UF
z>f=BzFd;Ocd@N%kvhfyVSKe&92`mYZ;6GDiRGr8seiJ&<qZCC-RL&YIOF%#AoszYV
zsDOT?>TA=bCANRXyX1N7GwpYtPM#zs(s#oZz))d_+hctPKO_KKbL6#n5cdnq#M`=A
z^-}w=1^M^HRN?|-p;M?mrm+16oNSsQ%~eOLZ-95`aAbh)BhR`9d5$WKEhgv}S^Wd`
z%5lLPgMEiKQk&@{`w&knUnrf`_A|@jV)X=hfsn|Tfdo^wI*Cr=x7*L8Kjov80k#PQ
z5ws}TQWYVB*{1f*=x6&yC2(0iW{T9Jwccnsah-|^K4BjP*AXkwxmFCEMqMWQTN)ee
zn)tNgVrycu)>*454#IwsZu2Y2WBdlB2~)>#BY%?x#D%Zmr_hzb*<dTMHx;3c;EVYy
z_!AQWw6qP>cIm5$%VIzog+8NW^qSV4G*!0B7N0hP-X*cz45B=gtrXctYW<;$wqx*m
zO-0@Cc``?I8MLnB8EL?uNT1X%%4y<j(TU&<tojDow%w9T#wcgld-zSN7L;O-WMV_p
z%m-;q{YE7yJ?!m~Ni-;N>T=|~tRZ%F0?``?1xLg}-h)KByc_$EucMd4+iZ~XQ22{<
z>TQg3?I;XE#&DANF7uUHnDItbtgVtg-s!N%{n>jL*iyRE>dTmnSfOeRT=I$O25Dx^
zFZ0pFMd%EIy2HLRN$;T=uA}}<-hEt-`!?6VWVX8+zW}}knvhgaG>F9Q<wqNCmyJLN
zv?pp&m31+uoj?_HV{R=^W{BcXB81L^-lOA`LijqCjpS~rWB_oEI08M-3`+js%|hym
zTLOZRjyTQ9d_DBfxG~Y6AxWAI6=84PQQ$P@zKM~_nFwJC^b!`ZL;lHd$HI=@v({+f
zG&IxQ5Fb~bol@?zDBqw%=-|>gsEa?U=nk3%lWbMqij5~Y<ce<-*siJBJ*in!rYz+M
z5S95v>RONycfpdwE#NxCT?5hP3Vg3vKe;*58?1#M;(BE6Z*g*GGIS*vmEi<FXP)Pz
z{QRA+sBXmJ((2~p;#Xxk^O{@jJ`=c7R1dsktz=qzh0>2aWYNK_U<rRsWbhX7xUw2b
zXa9OM=%>Wo;)C9v;B@a0%X*{`S%}R<AC}#7s|L;K7xWJ{+h5`Nv}2v?q3=7(B<%MF
zxdn60B_IoiGuL`QWc4gH0d1hw#xF?AE>i__#dRE9;cLi0E|t?e7i>nirR4;M6Lax6
z{~mb;JXpweMWU~P72I{C*q@SKvv_#HOVbo%6A(fU@<zcT9P{=CGMQ>7&Rg5t9%>H1
z0+R|G^IuBau|II94KCmz1eUAuNyT-!QGpz+(p!f;3A|?BSNsg6VXI1G;eOyHUz{+J
z>?|KNw5~3o(6>L}<qI*X*%$wEX1{x#mqI3j?@hDGQ9I^%rwO?U5aR^r14#`>F#D7S
z(i)_@bv>7n@`E^wod9-+Vxj9`A@b6+jhl|x*n7n9{)wVDFoUawW0r-Gru_)j0@|^P
zupM6S`{HLzH%)b|H8Ll;QDTJk1vmoTmv|q=O@$i=16P%}v}N2>%fCoFUrkrJPm%lj
zx+6f~xY=wGOZ@&>fv;c=f0VBuSdKQfyk-qj)%JF<8PUOHVgVo^OI)W4mqT|pP3C9q
zD1>`k$HiIj`eOA!FYu<&)%PGU5Zb2Nz#S%sp=OKXx_PIi{NSwx4@c+V?F01#Ev(s=
zd2)vo3$w!fr=<+(A@t4sRczC+lVzW$N&y3<h<km0_+F_$ZB|LWKqa&a8N+5-S_d8q
z_r)H<H_sE(*vwRPso2)hNZjD<FFaG*aX<1^#8zx3+d+D7c<dT5)38(A4klSz;N6?B
zgMX5iVNeffL*qRsejjoU+5k1+e&T+FvkNwvn+Yod&8(kx7NTW9L4&0kuRXW@U)iZ#
z6(I>}VVx`g5SS`faCz2Zdtam&9%+ssuA$!-7K1YSso@?p1__}N8GoYN%rAW1eG#P{
z(V#guqd~p%!e$^-j`Vi|>IW9HH^5rh0oOEi2%64Rg*zwlRvl6eZon>fm1#)mFZN}{
zu#E9YBVi2VWA<V1k-O{|VivH4DNDM+ofbM_Q=v$cu>-Q?2|In&@C0nF&;dwAmX>bw
zG{GB`U*i8TFHLnZZJ0U2+KiJXrqLqdy>AZo(A){!$F$pal3A*B0#)viSS%csudq|t
z2xKLC4PO(Wi!6bLKCCDQOlMs3LJUb89~c$Cj@bf?Oj!nOL+Y8YH+jT#5+4Hl*L_4P
znajq$N?}?781|HZwc)Ctj1$GQnAi-=e*kF|_t{!od6zL9eq>VENnjC@19XSZn!uu-
zfx+gjNV=&X)Pw5)oi@GX_s35&%?FqJBXPGkCwUYM%T(do#D-QHP4fOGP)wpPF3pO6
z@IOddV(Ms`M}83*V7c)&ZHqlbv&3cF?qRa800NR@l8^W?^I~|j?=zFeO!ohSMKLwZ
z<GgW!=hE%+)?yUg1bU8aONiiH{5Souv?=xwG6ruW97`L8Z^XtUuhOcRTCt+LD|Zj+
zB;MjWt5HB>@h#HZQXlA(xQXbU_SU!B+mY)?-s5jupO~@&9mL1DB8q{n++E=f{~>WC
z?=fvMjTig-vr2Z9ZUqtZkN&&B9kd%_%7n_DrT5Sw=8pwO*qidqguZ+OW*EC0{LYJb
zLts)Cf^0ytea#7H!UMJ`I?ix7%@N-U|1iH<btY3NsA$P{VNAjr<}fnQeOXexKcgR@
zF+Sb$GT{pH&Al;YP~0Z*apBQ`!3*+dfIk>+h<2hAIol#^w!p12Ln&Y2=E7_LJTD1Z
zQdfFMh>`yKz%=Am$#UszU?big>XOmEsA}>>=(KN&f1~vd{6e^gjX*Pj$AS;+>B;lv
zGG6qX`!yt5J7b+74qJpyF})<Bxr>|}_-2B*SmK%4_|7g1I5*`PV`Y0HAGsPOKln$X
zyL{DvKZRomEjZF!?fr#_!E3{9g@aJEDaKqAyblJUdj_NN9=i}I+EYz@r=IbjOxObr
z6{q?L*H@28{wwW)yDIcGttrsX{{mVA-ANp5DP%yijlalNvGgk3o9s3xW`0jR5MQCE
zk_Hg4uLXx&;#doDP#ThUG|&@uS^5N?5x>Cg(i40wkyqZ1EG>UD)$x@8hq-0UYAnTk
zLdn^P@&G#=I21P-9h$i)^FP9_{2KU0i2)x0Uy@ItBc-)sEnh_7Y#ijBfb<6!NkJsi
z)7cb_OwQa$Jm5MaUc5S#i+pdP(iR{g?pn!9<xlA{d_8WV!K#>H8c4qaE`TkFFxn~o
zEAOf|Fzu2r3gx^R=2j4mWQk_HE0U4%j)kne)Q10?-333|{t$1e)c4T%Oz%P8E3TD#
zhFffYhNXi~&^YEr+*!7Z^}MflOg~Y#lFSJ7FWD)yL`Y~d<dq%)xz$@1*H4(>ZwES3
zeNgVMV&5C<QACH^!)>|$y?>8^Ih*&#PPX>+&lev6JJ}Iyry+HqVi+^r;|Gf`dAC%n
z<GfzF0;<tCSI7Y}Q}zl5w<L8QWMUal8=#hDZ%RQ-3>bw;#?5&St}?2<Nzm`m;Np&K
zV`!Ustw_SZl>FzL?O!jLfbQaY^L^|$rjFkM_495n`m|xH{Z_*49j}3IfqZaepiqAa
zMqsnS5#$1B-u4v0of*meu9SLCily)x?=@(ivOJ@&|6r;hZ$OG6#D5kUk9=YKa3-{q
z99{B4%m$`ddV+~$qSs37L|hx|2O4wvxPyCaIS8E%)D=qjo<Nm`r@@~EfGaPpQq<A1
zK`ldBxvP04TmWqXl1%aDhQRmmTVb^6EI1hH!j#9~21jxW(AS0+C=xwMMj(H22LipZ
zk%caPxI%exKt~*=#6U)3GbAMU0LndI^EFe2WDr1Mli!5K2wjU(S;?~3-wt@>oy0wM
zMHwWk0B{6~K`=1_mie$!2fpZOmT^xysAme*rB|k@SO?!W{7PK5P_ec)_nzyH^k);?
zEunATJHlGNs?fX;4dih(anygJ1VX0;xTJaFKF*C@0<$t6l+FsIfQva8Iq3h@*b^N`
z67kmPZN57;O6ZbhDXHn{#k>|rSSthb%{Va7JhW(Iio-I^|556f=1Toyxsv{u<#I+f
z{vhgwB1&EL(R7dRccs1%fm8*%0e*HJ+6>yLgySE(k8@G{H|rZ|a$#d`G*Ekcxn~j#
zgCVgi8<VmF90L?sA42!U(ZYaK7@CPeMZ2J>yp!ncJIJoe`wC|ID^Z3$0Pf{oJBDs;
z#7qKFVIJ4Z_cO2%d79Qos^Rna23jV;U&SU!uJ5bN1u{*4AX~O>R!+;0!OhTh;Rc%G
zotH4d6r(11U+8&k8C1hu$ewpC1-f`^7<7~tSO?;nrGAnLM1-O6&%hOCp1DAYkajSy
z$x84ATq!ROTt`1zvOOX20~pD_P>LXz=P5Xgsn}ZND=|0rhh0CW-{-pd_VM}1O#~B4
z%wxE2;JjuxSRd~dE+Q-WN5DwybHSaGodkMIvBSVq#fBG&lMI!*$$!AtNHy9=<e)TN
z87OxV*6`8rQU3%W5+CU;7Kl3C`CKGP=u12nkGkF)D#3VI5A?N!GX@&d!Jn{<geBOM
zlzQONKtJGn(>)+fI?NA}9}$B>-vc|)H`pH`3-T>s5#+-DNCy7IHBvg8{zV9soJB0+
z1nU}fPik1$#6_B}nH!SjhC*pHG80|7dlYX<Z7k%Ac6Pd0h5HMhW7=38Q<}F;0oIlV
z@f+SYxwY9Jh^tH})5*3dcn?ezXJLE%eiG(ZagWO!X}x{D8avpl%t_Y4R=QJxqe`C1
z!s_wEm>yt?)=S9cI)hz(&wbs%T|zS<oU{k+VIG&xmk)Ho!f_-Ha$~)ueB<4+Im2tY
z9%v_aMaKGiXZRa`_8kB!ncnO<{WjF!QYNOD3&90Y8*4q}OQ<r{@M*HS>Ce62Q~=Qd
zO>m<Zp#`4HfiX~Bvy)G<{9}CsEH-t4K4AX|y_;6s`GW1qjzNZ6Ygv}C(}8xDL}gX6
zFKre4$~bSUFv8GuGf13&xZ$T62XrlbY)MqRGY$Mzd>7y+!d#<Q6hP5B%hxJZ^%XRF
zf;^RKdZ&1%c<Q=V@;8j9>@av*4T3p{o4M$l#<RZm93vhA8YYkRJxco`HSk;zPN$zn
zY9f0~BZ14peO3ZS!p*@A+5z?rJ_G^{YEEZ(Gq^o7$#<PEOv{jReg6Q*kS99_no5<=
zcv9(I;1{Vvc;_1kX?!E)2}~;snV#T5e|+Z8?tE7a`+)CD4CXILy@(N!x^Ud{%?Buh
z**kj%p@X=-jpG2q)7LvJP$G0lFq>`yPniWu4<b@HfU0~OsGUDYsK$ND{4KP@`hu-1
zpM)DhN0^~ZBHT5mP5mQmSU74fCYlvB<i}ga0SA*M%UJQOSMeUu1$MJRYwG3Mzpe-K
zl!-LEe8W+W?TNku9|$~V2cNq~t0$31u7l<g)|tpyqdNHvpL8AZElK||a7`{y$;<}G
zH1{fW8n6^7fGk0~bPX%dyzMI&R~LUa50_?h7fauYVLsn<J<$Zt0^+43#c7@`-dZS&
zuJb)%pB5i2a+rEC-=OQvUd9HtR#v9pH25&LOnL5nHrHzhDiiLc-Sf4Fox<U`N?%w=
zVEV})3#Y}_QGC*R;Xk%MIy7yG=M{^yK8p^IOOMS2;o5T3rq7J-xV`WuV<lThG$SXW
zrt#mKuVxg;XG^<4acO|}4zLK6m==MWfn`S3vneIfTa8(sS_9p|;D8y9^&{RTOn@u(
zu4+;}LkV1B_gW4oPEKuZzUUq0iO87CWP64~d1=c5iz(XL7SgS2GC{C+;ryNR@>iAm
z;kj&ga0<6P(9iVNJa5|vb2FfwsrR-<*85)04H{KvDwhx!o$+;N1(RZmHO)m%FmY%c
zk`-Sfondc#_Xeyk!?9L)jvJoV0vO~g_NB0gGI9+U?JMpIo40LL!Ab@eHu<YTBho%0
zx4h+GFBj)+Vlcah086BamIz;5eA~=g&^KXA+B1KAa{J<w{zYIl<f8AeaMbeLmHS%{
zV7l2FILM`W_ixLMtqcS+W_mHpICQ8`ESxm9SV@LXxgzsg%xY*Zlj4c>M#nD%`uRVV
z#BdEt3b?N(i>VOy21sV7f2g$CdIL%a&&OCyV~kVKGQV+KVk)6-mi}lv(<``1Ad1)1
z5;J!|*E}7Vuj$1Sy|qfMeQc5WD^oXPzty;X)|bFNtCs9A;p`QEc5#=|@!Vfrk?)On
zlHqFY4Se!By-9#OFw3`_5iL7RG;hne6W7s?nHX*g^dyka{>)8avw@dxF=?RpPF#Jg
zX2wtcQAS6&&T>;Y4R5i`W^${<8lL5qLZvUpUj$q$ejc-Ko1Q)?X+>r{(;ot%OlZWW
zDzTeVf6ExZ;};9Tqd=~o;wW$*yUMd7DaY&a@A6J%!u~2iZ{J+^GS2|0x;xiXV9Ft~
zEMwyH6CU|0(=&u^%qib^)8*t3;AeKJsdmaOc4y*Swk_}lZp&H#X6uVo0s4jA%G}D_
zU^>9K8aaIr0jO}UcM#%9n-@34@{9EevdLQ2RL^hokM*zjzT(z;JFp9U{nK*`E?~RN
z-GB%1o$ZbMmbi6r8B@rTmPO3Lj23b4Jhq}*rroXr?<!wCakSyI&Jldp1b&N&a#i~7
z@S`jf;;&{74Gc*Cm5oVVALxOlc~AJJfp@LrT`N%DmCyeNRQi9-Xqi4PZYY;v)e8~@
z-dB=w!}O7-w$A6Crq_3OMBZAKdA3+4WY+Tgk=D>Zg{?gs`77Y7yp6FNO2LE?%<tgF
zk_tG@ypVeWS9KNpW5L5o>$!o#R`+0Y18^;~ksImGNPq%)=0)ZQ-onPCtnba)X0GTM
zb0`V0yae)1m4Trtt(YUgF?g++fK%NjxF~(H_`p2TpIqF#@q*%}CC`&0y}e|Ue+7R5
z7y}&N+&lHW<*xe}JHXun>76{U#KaV*?gv`?yy=YfM`UHvNwcFQ$%iv(4KAC8rmgTj
zH~(vS3ip;XlLy#%YN~u1-NK|RSBO34uB6!)Vv1~M4EFjWaV98g4Qy+596zYO6`Z9Z
zIZEk;HD-Px#o$QW8J)+x!1_RRutvgAM}@+=qVYvSZ%@3VF;d0T9*$1?1z#w$fl^sg
zddbZl+cF1{@3CuJqlCI>tUAl@QlBHy^nE0?BBHDnrDPTe)z#+WXYIW<$}Zrx&~oM^
znM2OoRgBlhjw6>!pTu1z6145$Ec+CHKGBvC)M~(m!VpnE*j>>|CB2iq7V?Yjq)-iO
zNUTS{N4_Iowd|=R7(4R4?4J^<*5e+q_poYd^N@Pd8t-DIp8YKPn(v8=1e(>8I;Jks
z%WP|4)Ns6|CL{Do$6Cj9VhT1*xEb1y27`wwNgbwc4-#r&t8(&%cpooQyV>%+2!TsE
z_@B%Yt(&brKa?A8TW$L+{v<6Yr*SLsdBOJHsl@b9l-dGy(HU9aaqSFs#1-53nu)$o
z`lxI@gLx;d;l@FL$sC$Z<x;n)Kjn>;mxy8Zf~4A{>{uxMTUMV8!sD5V;ze6+VwUM0
zb;2^7IxS~gU2=ue%uj25g@W|j=zWc_HzR-6+t_QOAbpV;ruPZ;4?B{dIK+(KgGZDc
zpr$o4{eW($vAlMrMDk*%xd&+w1q!arGTUPQRMKG0mC%_?kv=D}q|3G?ZcmDiu2ZVo
zCDKA9FfeTvv-N4gyTJk}R=tRgx2@wEVTK|}xlmq8kCK*oj);TBi3A>esPqDR7%Hwc
z)C-S^?xA$0>LJaH4aHP-9QhZr6y6uAl%2%r5?bg%9^P`-sDm|ock-C77;2}-L;~#z
zE_W@G?*!LKeJF|Ti;py?g&Alu@v!_4_$W1>+o}IS?<pOJl!P|n=^0z-G1`-iG}K80
zU`Ki{!a`kyhBT)9m@zca6#EeDqgUY*!A`b0+7C)SBqw#3?#CU{L*NW{Ic=3XgFyT<
zLoRQWoXEW*YbeR~#!Z*NmFVBf6tQt=JNYv<NSb2uivluM>Tapj)2UzaiwRj|kWZ4b
zp>f1Qq64W(&6JCl7;H0D2ie0#Nh@qqNgeBs4AE7*CEYmH$-YOBP=7pD8Er@SpXvF;
zF?1vPK*&M%_!_`8k!BwO%ts}UC=}X*YIlQ)@<qd>Nkl{ZF}N}Cy$~(cW1dKv;-=8R
zJ^g7PnnZo|_Ruc{?HZ#^QI4y%kf!=zswLW8S!bCO8tktv9SzlIS^|xzWpty^&9r08
zCGRPFTic^Ro-)R^CiqQ&q&LYu)ki?b(5w`z{cnAtGE?mYjWD{!W=0L0gn;@1v7DA2
z=h*%J*UCX?sFp%}PO78svGpWQ606lU*j3Q1JTY7u7F#9KhTLu%M>GhIHJH6;<<H?q
zw2s-8-{{%O2=Th*qvi}g0yBlL+zKBp%@0-C9--#gAD6F{pQAOUc8*f!c`!-oBu%xC
z(q%j~^@Q@pS!piMdaR}e8(E9-KW$cg6FWm$OUEMLJcFnQu2Qu_&?Ue183`-KT%`>x
zi75(8?6nW3KVl#~&wEAx1fIkG)Mxo;=$AtaDw}9+%BK?BHH*4d-lT0$-^nf{qmwSm
zAGj2<5}j}BA-7h}hH`<Sz9#rC>}SjGsdEEK_zN%$FQv8y<De$`Lp0Agm2`QCrz@G|
zXh18(jnaGC9q?6fmRd=y78|0xxXpA1Tj<-tTs1zw($Nd33blhy8<glFz9yjn)tI@e
zIChRSddga%8qgK$f>f9Cpi{$dZLJl9>!8B=W-?CM>s`tRNT2PwttZo3I18DSOl$^p
zL3wPeVN7<9;bWB(Y?yxPKP^?H77**qGsKJP0oz=uAL=S;2i0`E)f<qw+`yfrHN>|n
z+2Emyy+S(aPzU>ls-%)hul2sg;_X3WTltec)qYX9sjQ<`;ghf|A<F^>mtb|B&!~89
zIMqvfRNjs}j4Z;2@)v`PprOo6;XC^#d=7P7JAs!thal(G8Oji~syt4(fIiS(0lxqg
zdBkklo54+zm(dj5P(;`t<$U^2ay5Az2nm<0y@M}XY!iaY7w8!94?o#907y37RQjT1
zhGMQJf^0RhI21{oK)1^gcs@6ms!cQzx_Ecn=F=18+4?uKq0$sGdXmsFptjgt{|@>@
zTvA=+2^`i2mcMKMD0l!(<fG(e?mJ>@Ruxm!jl@^wEIAtQ=s&3RNWbjWG9HL|WIeKn
zQV@TcUM>F!P1MB9iMCt5Bsz|qsLwR^^5dM7@JHHyqGPZo>Bl!i$Kf$(v9ek{Vi3>Y
z@m}g1`ZM&FNH>ikt|$E+Oc!5=j#?LSl<jo$n#vzQecKi8MeqqaQ$3-k5gw?K+Jd<n
zeyIJ;Ez^dpPt*&hzDgJAvK)ck46Z;6^w)T2#Ldk>E=u3ociKOs;PiC-VzJ$zfL8iD
z8EIVYusu7EaHEyu)LFH1`zqoR&_|kK2B;W=(R|!GMrlqun>(pvr7`xeVkalB-_&wr
zt@>BJQr+juwJBPR+8M~er=xR&RfOF{U%ZF3k3AWjXKYM+5t_}V|ESvNDNInT2(=R$
zpkYKM`kG#{E>59_!~5{M!8RTfoTb+VJIIo4j>VwZG25k9^yau!IttrIN3b=33cO_3
zO6sv##PydXR3~pX1#R7I3BU!yWb(4@ZQo&0Xc^_9FOg$yklI0<O2x@P2t{-o<JYNW
z<lNj-he9$tOa6{(hVPN;5Kl9POYMlKa1U&@>h)i8l9?09k9s?05^~7lL6W2=!q0?6
zEh$S1?obMqrF6bHz4)OWB3jVfY1-yuT7;<Jcz<2&Z*3@Ey@GQZY!^ZTI~n`fiyoiy
z2+);2@^WM}F#xOLx`Pi;ejp|ZhcjHt0c{MHD1768qR$I|p#Rtzbz5)~)gn2T_aKLY
zz4-@Nk_r<!-o}o){!gSu;-spfg;}Z8bUHu_!H$@nE~sc_5Lf4zCdB00Qo;+sC76@m
z)}n|R!sO6%v3zQX;3OVuugwv(q_4Hy)UUF&y>-QM*I>tbyx7}Us2iM#O~iWRF}y2l
z1C)mRjUCXt`{K2>Fsw_=*8qq)fJOQOaQ`}y9;i6=c5*CT)Am!K0IRC~uJ0+t;rc1L
zTB&(#<^+9UdMfuPmPmgp+$Bwf6STVNlkq71jcoz)2RnxR$9~&}Q-dv+k;O!-@MQrH
zJ$5HrS`i(Z{)2DVzvsUplk79>exj}RAAFj4E`0I5K?PwVd)xdf{+RX}>42OkZd*(E
z{eTDl65Q>dq8>yV3omzq$al0sUxcel!~I8t@#IVPvRdBQhm3Lzx8z9^!?V1f&>Ve>
zoPneI27iPePcAfV$DA^43yU3$oEMLtYF~@mxsC!aI*^)%YH@$$R%Z*ohIoqVY226+
zKSzy`suE{~<w&&cK3<2&Do)gEKryg{4JrL;SH4gNttN(WHS~_!&U6QTL0HXo!8@fE
zCfJaPq))HH_6*$y{^etZ@yJ7H0x|%%hQBM_LM4Qc7agKvO&T*VomZRc9r1Jig{lMZ
zLu*=+J|BvflY9Yak@+b3RZ(qCX(LZdJC&rSH?q}|9}1hn+wyvoh>Q9Zq#kAk#hsdM
zp446Z4ve7P>Oo>EzRmd&Z%+M0Ojj0I+M74oI_dw4{}^<@dg9q2rAOmWL)*e4)=NDZ
zniiy~JxXn5k+ROHX=mv5kzCUPWl)R1k$;I}MqM_7*(SH9YbxuDCSeDKfyN~KKl@TY
z#*Z{NSL+h<kx%@A;__gxU|lqe92&Z=oDr%?q?Uq`Msf?$I5w?w+FR0sv`~^mx9R`%
z=UPkfrd|Xb%<Q6+YIAid)K}h;Woh!AcY#|}ywaLdP9c;xbZ_RBx<M;Pnq@5`VZD!1
zOZkLUMK5YUsw0pJ{e4y>y&~L0j8m)FS6CO>mXHI}j>(-o50pY-2Oa_as1E>6n&z2@
z=($>3e!pJZ*|~{LXs1jzpSN|Dol>7rK<Y{TYj=6F)qj-ZnOxwV+RJ|5(l)a>qNvHp
zeI<?SskpfW*&&T1Z`kiBvr#TM3tb@n8Dz0Zfsg*ejF0G3d{SjEc@Kh0TQi>$KjIj1
zH|QkK$<^^I92ICrO7Tx*L&ueQ)mk5ZsJ4>VXq-b+_H!)xD>l^dC?5{L0Ad}r0x`rX
z?T+w>*r;0UW5n%fGjRuwNj7|Y78(qxq)s@>)ua6U%p~QLGKPNO+vQl+%B2}wQ<7#n
zD-$UO1jvi%eRCBw3G^7r5o3LDoCUeb-SFnt8W_Hnwu)F_?9lq+D}WMl6mg=wH9i(2
z!3heZ4!19}uhA+EopLVu#S&Kgds^FE$`90CgJBn|29)DkmSeZ0s`fPWFM9<#K;|(C
z%@V+V{C@JjZHaQ$(_E{nRw&tOoO%i!z~%UQkjKe5_@c5>xMQ0f>Rr|zi_+RwL}hhS
z`<UF+COVk%9A1vkqz_Urhzj90wK2TTo-W>%UI{QU482c(D*eRw3r4$Z>n+G5a)dn!
z-AA0V4PuJ5c#T!&Q1$sdwTXTm&Q`~VQk7MuweybyTU}S^|78$N5uPa&@kU*(91y;#
zthC?s!5t0%h#|@bbpbs*#EW~SB`u~TomFbXJK3j&%d{8xTzi&RrT5sE1d|&5=sv&_
z*k~<A`4=A}o{@JFH@q{`xj+}8zaFj9lqlp9Z&DxAp8|(X4ZsS;t>x*Y5{aHri}aYx
z*sLdH3cz4_)N9Zjs-b}>Pc_G;0#((dz--Kq{T$3Dsw-89&De8#hG50|!q>r@NQN@S
z-j*JZ@&0O|!-=Vs6TEBRhg5OZ!~yDU@Hx3pTAn^2e2z=a;K<*6AB7bds<kd_Za<ls
zM4T^d6UawT62pX#CD$#I+RT2RS!TO#q?VPJ)gspk&4@Okdx7~fDJE$X<V!M&By+&7
z0~oFkN3WBRUX%EQZbTsE6lM!Wc^hKUVr_#U6SE`BzCt)iyhS^U2ekWm9?=5%g!bO@
z1Eu;L;t0b(H#XQeqqFIvx;6a0;py-oPQkbU=yEOF|4XR3xEk%7WfG&s5}^($>x<)K
zkfpZg<R8>+fCn~HDsrc0ZPMz>5e&MoIEejK2&#Qesn`Ufnc)Bk;Ol%Jlx=Dhaf`eu
zU9#suSCDe(yNnUY3(rct)F5yz1Q2mnV5YW|H&aRCBX)HuAuFZNr9n!f8xibk7hg9c
znP{qfPZa#u<ZJR;VLkL(ij=n?8mUv+z(wMWxd41*ZBJ}dc8Remw@lxly=pCaoa<Q0
zcSb_Y4^padn2NW>D_?^?&pRf_&NI>j2OBqbkxFI8syZ4HPpi(7{_F;_0ntiXFPt$5
zYQv-)A_7m8^Z4Fs9x|e!E~N{f;^URBV%5yqN`F2A*~?yqXUSiDyU}rM4)n<XPW=z@
z$S$!3{*YSF+(~8N_EZOn#G3PyiFw9Ed<R&}ToE-cQ7XevVueB<<T`S|e_1Y+uLU79
zW*d|FHxCBx@!RQh;xYLfnxZcQ)@XTwgWwmTp0pTyi%xWnF48#;+iq))zX|Lod91Fy
zOgyRlntI(_F1~k1m|qLG(CvY2p;Gvqo&wx3wxT0st88Q(`Ti913u4GWz>kUtEC|lW
z8VDicwn5D3gf~iDLqz$$Db<T-DeaZFJKYpyiLn((V~D+IKKr)>Vn3U9?1+SX{@cF3
zwwh#L+vdXGQLWenbx7E+=2*sC973uD$_LOIz&T5EgS+&Km>76x-6ffAlZZ=LB)h{%
z3<@Q$!VW7F-T8qPLN5HPxHR^QWRdFl<|qw}w2rIFMj}IOtp@BXpwVcqmz9(TyOj3Q
zH5XxcsUl6wg)5LN5XS!sK6fuhnyO2Et+8bQgQoZjfis}!9dAL%op@bxq})9%74Jp(
z0-yM5#;kG@0!lBC+Q>WX9`^_wrDPQMjH{x)i{->|@?7y#pa&2xy(T#MW=K1&x-ebT
zB?>*Kcj6b2cY_hA32i0*Yp%gNj45@qWQbV6oI`%HK2f6s$@osRrfsn`DBts(HBTs%
z3NPSKeY1&$`7aci%k}oLq+y@5Sy;ZpYl+|lY5@{q{e_qSJrU31hMbZ7m$i%Wp`J<x
z>G8|jOW}9Lm+^PTR-vvsm-|31<y(v2X;ChZFiR`O3Xn<4xs02cWf~w3QdejTq4n}5
z;xD|Z_*8hu<9bi?EWsqB!MS)b)+pscdJl4$kr{ZNyQdYY6NxIEM_3{j3U$N?syFcx
zeZoc~0|-{q(Gu!cpOShO*#ca{7E?O=2y9?ULK-4ltUmX3VluT%|6OjD$s=EY_i!_H
z4j+l%BkBcz6UQiTlE$TP;7=oep<!!vB^g~{x=p^|MiNoRE^#2|5Eg2XsknGHd>QRy
z%}yqQX4`)0OM*KyNG){lPIFPw${cJFJfaau9wo=46~3zIEqsxFI9N;_6O!%omFB6Z
z@fIi~EktsdZ;Dx6VQ>f{r0l>DcN=Lv+OuhF@g-v`I1cNCepF-0x<VJ?0@>Vqji>?q
znAe5#k*jUf&~9W5a!f15dm{1V1XjR5V;;&1pOmZEhSL+-14gDpy21JF5}w2SgGNhN
zg}t^W_<mz1-v+9UJS3)q!-UCr2S$KbgV!XOm?NBI@|A`D8jj{zgQR!JFnpxYJV+r)
z*6WnRyFjc8bOuz%CtJ@fq9m5QELFwU@P#rX{)DbWnow_59%S`1<R|G0+(01WTftYv
z)BXw4T%A%Ql;^-J*zam>UeE6hY-iWn<J6Bzsz#|>5*Laev8800d`66MJ!eRBB>V)c
zt^9{}rEVP<M7#;cpbf0O=@ablR5KO;1$Pd;m^~uA1g;=`$eSpbyw|j!Sq9qB3*sc;
z7F8wlf&8ESptQl*&i{r!5h&$`?|RacsJZ9`rK&PX9woQa3WY~92TZo@#5QM)OCIB!
zY95U_xM}D?-r#Z*uc?ppF7$(59ht7o;-YA?Fr7GMeTY4V3XvuvAaz7^-&pmmwnqEp
z)(Io+HV{R07EXhT|2jH2*%BCsS0f{=vqVp5Rbm42Se-yOOn(*J#c4`Dnj@^0dWz8g
zhRXNYEb%d44$sw2<89#*@|)HXzh_>m96`Q;wec&v8T(RTvL1)rhKa<L!UNgJE=*d7
zZVI2%>MEp=UbL0o<{VDU!V1&IkfK@{aEs%SzDj23n`;c8Os-VFxNWZRt}m93p<Uuw
ze>8fxbg^msE;ZEBNMD-F{w3Ss3gs_#DEkCKk;dXEayer5YYH!1vv#FA_y+6c>PGD<
zP}e;of0|eZ%k+67!tdJ9*%wbu<n9uErCLONeXFpp_&v2i<@ri?mgN%Kfn5q#L0#fE
zvI)t%RQYdt0Cr3`0DZ&z^O<55ak7Uje9Ih8-9n5oH6v0|?~2!ip!7F*$5fG76FzJ9
zn!c%SWsczok5#K9ChU1hNF1(QRGR5MqyT?$^D-jFdY@>8-oxsESsapa-`q#4W^5il
zF?&3?zZP;8-A!D>{w4C{DcCyozu-`;!@+4(Q{cJV!#jd$>LjHLHrjZ94bTRW(`a4H
zc2x$dA~a-9t*I<W|HYS;mm?2@>-~h*2AOA@2Cp{jsn?K0+D$B4xx|ejM*HKqcwxGK
z2z(T>p^rt(^B$@Wr2>y&L$w2p5|Xnjpg{x@Ocjj;M`6BE$MjME3T7$>eU-+D^TZ!|
zjI>UU!V2+R^|-((BdKXbZ()$U24BeKE2^}YEwYkitooY118B@TESY>j<_Dcfq-IrG
z@88cP6Pv|s(_ZZhz777_HBb@t*Mgm{X?X4@g14p5wttPTc?tY1R2U$%n}vTeP>DsZ
zf!~6rh0uaU@MrU^|JRoN2}EICrBwmDit(u-l|ZSd)|%js!jVKB`$}j^)7oSkgMd9+
z$;UdYE#jl_BDe%S>bhmnyd%ZS@W1d{d3p3$xGKLkWi{~9mW%r^&d}bT#om_~+Z+W^
z64)wm7JNuJ%GZ={aSMg-fi~En2H&tQR4Z|;bw6^^Kt|284AY*PL7|q^2>eK7kuN#6
z<N-5PuFj8Brb33wo!yRJR3f==Sh?$?|DjyNdqZl=O@n43`>FA2Cfh`EOOY9Nw5jPb
zQV&P*-g4)-bHE6HTYm)-8@z}&(z1P*;5b8H!C8J&{&8;-kC*-m)YlG5$=YeGH948+
zu8gpk1aVG==a~b+(cbL_bFhHRRHtM2=sto=2)kg?qYGR#cr0^V_>FZCS|Cqzk3h16
zYLJlTkXu4u)vE9ixl2YAd4by~W)k)AA#6i&B7H<oMNVa&^sYpPXiZIDi=y!oqarVc
zA`uI5n{19(3s&(LqZLLMehJPak__TU1U1XFfqn0Z^LGWT1i&`f(pjl12IOasn4-Dl
zF~e=lArv_sehK>d;~Ht&N1Ze!lkJdA!W|$VUWq?NS4hd!AhnS(v7E0ivDD|*%R9{V
zkfzv2Vk@$qnTCEXr+`7?E93|8jd_v(O55j-h3F?Zg`H1*WPamQh)%+2JYd?3ZBR0Z
z^WHzvhti$=k+u$Ofp>*3A@D8Gmh4D)rPkVc;H=Qd%VRwJQq6ZS#x98!+G%6`vx$${
zSp$wCI-y@JhvARuv05A+ttOIRs66iv#*QV4_({k?L&7OG9_#Otd|`T;yCE}QSP%Q5
zCd#ggL}d%s1$|ojxA~d;TE#&NbC+-d9kUwYdC(tfGh~5qo1m1n+JEFTl=NS>U$#{x
zZlSt7#@0r8hAtz4&1?K$nWkuzSk4yE7XXF&Db*^y;Tj1>#Z=l+9R>b{KR1sSUs7Sf
zf;6YTiVxT_<5NfFUUXluoADI6wf0QVYNHkbI5ms?Sy>=;(Z*ZS!5jE{-v#I#I#U{s
zz6XuOAs`>0?VGK>CxRJ+Xc8QT#&Y@OtLojL3_>T`sLSKVgCEfK9Ir(S6Qp?Id!mv)
z2hZ`}B)SXx*=g&)hNlRf)p73U{)53ia)h!$Di5BMpESB@@PJ2ZwdGK7q?Vq}o7+S6
zIL2$oY-$(eFJK~k4dm2Up8H}A#H>%&CXv;_)kexg9efT3W2?mO)->uf`N}(8io<Kj
z$?{tzL@m~u_#EhTw6&*7kP@3?i`04Uq?pJ09BP9&P5ps=3D@60D)cGgE4GsyYV5Ky
zb;CuATtoY!ijINn<4-7gfivb#fzedo#5%$P{k!D9kP!k9d<i}ZBBi=si%RE`byj`I
z){qa8QOZ?iXV8RrC6BNk?<w5j2HMZr{6?N&np$0XX?mwzRo5_2u+NxoD6jwV^>9W2
z`SK8GK6#W#fI9i&=%2XJ_$s0dQuw=}W6UP-IPn4D4PwG5vAcA{n7Dm!T8o&G?+xPQ
zE~sP1zwkuINVJ;Ws-C9b8g9y3#P>$}o{eb`tPm%}rUO$-!s=?WQF?FJMXeAm^*51n
zp%sCB#D3F8=2xME8YfRx=K=Gm<K%ATp^Pg@0D<)LouW;eU1*}ags$dp`g<eSWvd!X
zjmfMa{!q$<b5b4rlkq=3*?kewv1}vrYXk7Dc#`QBJ06*!l@Ob)^Vr-CRcUYO54?+u
z#-6(;@*DoIp*!)5s%ql^?rSK4M2NV7G6S>S_1t9!!zCb6(>^7nyh>WVpk|aCeiB4e
zOG+IPvCs?xWoDS=&YioQduISqG%Z}h%!C%VH%T*5%O&;I%IH0R#QB`(Jm>j6-(XwD
zM)@nESe&30(MSAgYIX7-E;F_}fPpaa1b*3mk`YLABp)U!9df>!-;fRB^$crO%Me8d
ze~VB$uRH|n^i=*d@_(k8YRJT<&$NiDnJ8O1aTy*4FF1hfHTOb%7_5}0gE}}<v-4-c
zaOq|Jfmo@Ha7>gE*gd}cGUTr_1#&hyG#t(Lhkn{Sm%GN<wFk^ExK+yz6n#{ab}LxO
zcZGi45e1#VNiMSC$f+Kz?vno_42D;cU2#{7SEI~Hlu2&Xo)OS?Gi2p#a1f)T#EsN-
z=}FEi17F0w%z-Iwx(~)NrybRFB|k$Ns&@Hi275rS)F94L_JKIY?uhcPWzyKq$}Gj8
z(Rzi3^4i3CGtc~s94u_{63Q@szVy4c-t;v)2bA)^HZPQLco}qtT7oK9!=U#*ZkGO3
z22$PRVtSI07U_}vWE-YPcCD}FAzo})$NVnGCOf(DnnA1dC(FDDg?4sm>t1p|i+Ajc
z(x?2Tx<%}Eb#dgjKQ-f|_8~b*BP`v4t!g!v8+vK{DV0Rsa=S`MMn?np4cl#Y!Wr<6
zzZ2WVmgr-dS=us_m-suYpz%y<hTV;_RWsG=%D?$RDC7N?JhE;9v%oSHe5MuRtHt^3
zXuZ%nkN*M;7+U2}wSN0tstoB1nN&6M$DRmcjvw)UOQnC7UdLw|irK#+zp!s0^U)9b
zn627#l#R`jqRIPGYOs}yVct}3<5M()5N4**V$&KdN$KGfM#CP!Fr6x|5z5$wLJ_iO
zC(tAOL-swwcZpN!juxt|+{1H^r*)`7DjQAvymzn_PS-BL0i$`L8%)qQ@n(5B-D5Y7
z<dhDka9e)UdAdk|s-Waa3=7RTy9M-`KlCmZmvBqOxbZG#gYq182uAD4NN($ptCXUK
zo3u>Ur+yyvxQ~VRvw5TwTo-4`wc<`kKl6?lZ>f{wz_)M`w7R!(xm-W!%6J*|;QCCD
zpqX*duc2Rp4*3!hYkegpHa(BdwzMMqD1cvRxJtz8tMyLnvbL4%biR!3-aUuPar7Ia
zV5=*Q?iYgWZMFmHZUq{vi{NYOE-MDIwS#^$tP5Ce8|1Z38TRp7bR@^~I<(1mFkN{Z
z-$>PXFTxkewZsyt+ISOeA+B&9>lpcaHo^M<Kc<dgE5MSB+mTmVjId0#DUH^4v?ul#
z#~N}ycmS6=5=fHHHfGpIg(oYI6SK)~<zetob_YBcFtx0y$6OZmG{bpc5e#~PIS(F$
z!-t;=Q1HKUDSkKOIGTdL%^u*>)eOh0{6NPQBOfTX>KdOhMQYPFQ9JgGRGrvcpiDX>
zydjU1+{#cjS$sQkMmQ(=^b5#(-y0r9?xYKua=Y8o4<^u8?bTGDxkh~3an%~n=Sc<f
zS=gAi0?g%}l}xNL;Nw2Q$2q2wx0=pLC*d^i5RqURqPg+<2m=-$2&4Yd32umo(yKVe
zctTE*cCw|C71p{_wHi3nj(z}}v-pGh9cGn$0<3BYa09k%px!sVYA*4N(QarvSYS^e
zI`&>yH)-oFG03tru@C&Sm1>x5>fO?8XQ+=rS@?*S!?LVXa%oGzC0jQf72KxY7fQzz
zqB~g(#h{E*zI~kV5_gxN8gjPBxx(&Y;S{1I)ZpX1WlCQA4E6`Sn*52LC|wE1NdL1|
z>UHuLXy$Y~>?E^c1^E@SKqkUE@jEt_oGlGPf-M7+scNt?JRLt6ei+zBZ6%g7HP&6m
z%j`Psnv{vM<TcE>uxrl=|DQrZ{buuw%x#t)63Ma@qZY9@xDNdQ)yovpCy|qqMbA@e
zZO_xIaOj(_%!LE_Rj9UQHAShEH%Ld6dL`Am&uWtroENQsFcs<|x8HYIto45Fv%~M*
z0cw+4z%GP&^j@`?K1o(1QF9G3EopxGH`H;gSiQ!6Nlm3&a2b_8@9*9xen+L?Nz#+@
zJt=>r!MjY(rAP3SgOhCO%3bqF&p=bPEljT_c-3N?>@6XkJB#oSuwqFF4AKHp9~P7w
zn|p=V<*24N>Ys0_ln1MKK$QAaV=6}@jdCe?mMY`2)2h&I^?|1~^g&ZQ4#MxaU+|Y9
zo1>H(hg%zC4MP-h?7LFDFqex<?qCPn&Nru+hBbb!PSlUFZflXe&Yvr$$;YkFc(+r4
zjq%RG*Tbpc6nVg3kLAHbmd}aTm9jB+aM!k*uuDtV-qSy2nt}`Qhag9ZHkH)2*Yi@6
zWt#bvF;RFY@Rstne>K(Y9+$QldV<JNfgEh7pR&XevsFy${g@q9%pOn5;E!7mDJzs>
zymf3N713@|0W4Zgr{|@Oq1&ij@X&M^7^yC|Mmx!y!XvKK7C}=gD}s}y!In~SDcxsT
z5Jdets!J&M%nr2ElZi>`3AMXIXHn9pswsx8kQ~OfLXGthyd(Z00!H;FDaV^N>9C&Y
zFAe@<yX)JDmy>%!A=<^PU_T4A(50by@E_QRw%5OfFC$lu%9J~IU8ddEiL{K>Qj?VR
z#1wB8s!Big+yH_X?Z=a@8soJq%nA7d{;qij`81=eB>B8;CfrOPt$QCo%QB&T&A*91
z6NX5S5HTlWr@f0znbLAI;S%IWb)A;2(ofPSWFi)8zDG{*4yTsWy>hO)8Sl4RpkHnf
zp94=2Yl)vtJ=mqxI<o=Gvka35u+{#W0HOQ^&PhXv$3hppG1M_nilLmo0LE~x#B<c`
zrnSrw;YoZ8w-PREzGeQ=|E#%)nE>}t2~-v{D45HyWX@1GnEhBZ60W|$ox*Tw3Ng{w
z19Q}(k1IlFm>%#Iy#T#r8Tud7o3;k#7<M1^h__m9S&GTQSPl~SoWXMaMY$Hs<~E75
zOg3+7XtUg6``cc@=7dHdiS$<3&Fm+-&ErFvu0hQUsLI{%IXly;Lr2Z)g1=!GP5ZH>
z^wW5}@GYGRN^BRY?}RRQQEC#bac6*A=~u>nW6<yd+f#Iw*AmJPWeKX{qJALq!Q{wh
zs)ga;OY%pkllPc3+LJ_V40X_7D)+>rKo9Pf?|F_=`Ruh|I`*x_Se0Vwhv|M!L+9N;
P3Yl!{g%Twjp7H%3hOI(i

literal 0
HcmV?d00001

diff --git a/wasted/gtable_to_json.py b/wasted/gtable_to_json.py
new file mode 100644
index 0000000000..35282dd8a4
--- /dev/null
+++ b/wasted/gtable_to_json.py
@@ -0,0 +1,18 @@
+with open("sheet.csv", "r") as f:
+    lines = f.readlines()
+
+keys = lines[0].strip().split(",")
+dd = {}
+print(keys)
+for line in lines[1:]:
+    values = line.strip().split(",")
+    assert len(values) == len(keys)
+    model_name = values[0]
+    dd[model_name] = {}
+    for k, v in zip(keys[1:], values[1:]):
+        dd[model_name][k] = v
+
+import json
+
+with open("model_info.json", "w+") as f:
+    json.dump(dd, f)
diff --git a/wasted/load_np.py b/wasted/load_np.py
new file mode 100644
index 0000000000..7b92a7626e
--- /dev/null
+++ b/wasted/load_np.py
@@ -0,0 +1,56 @@
+import numpy as np
+
+datas = []
+pre_x = None
+
+
+def concat_qkv(datas):
+    qs = []
+    ks = []
+    vs = []
+    for d in datas:
+        print(d.shape)
+        qkv = np.split(d, 3, axis=-1)
+        q = qkv[0]
+        k = qkv[1]
+        v = qkv[2]
+        qs.append(q)
+        ks.append(k)
+        vs.append(v)
+
+    result = qs + ks + vs
+    return np.concatenate(result, -1)
+
+
+for i in range(4):
+    x = np.load(f"disco_test/{i}.npy")
+    datas.append(x)
+    sum_result = np.sum(datas, axis=0)
+    mean_result = np.mean(datas, axis=0)
+    # concat_result = concat_qkv(datas)
+    concat_result = np.concatenate(datas, axis=-1)
+
+    # print(i)
+    # print(x)
+    # print("====")
+    # if pre_x is None or not np.allclose(x, pre_x):
+    #     print(i)
+    #     print(x)
+    pre_x = x
+
+std_x = np.load("disco_test/single.npy")
+
+
+result = sum_result
+# print(datas[0])
+# print(sum_result)
+# print(concat_result)
+# print(concat_result.shape)
+# print(std_x.shape)
+print(result)
+print(std_x)
+# print(concat_result - std_x)
+print(np.max(result - std_x))
+# print(np.sum((concat_result - std_x) > 0))
+# print(np.argmax(concat_result - std_x))
+print(np.allclose(result, std_x, atol=1e-1))
diff --git a/wasted/model_info.json b/wasted/model_info.json
new file mode 100644
index 0000000000..6d2f7409ea
--- /dev/null
+++ b/wasted/model_info.json
@@ -0,0 +1 @@
+{"Llama-2-7b-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-7b-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/Llama-2-7b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-7b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-7b-hf-q4f32_1-MLC"}, "Llama-2-13b-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-13b-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/Llama-2-13b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-13b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-13b-hf-q4f32_1-MLC"}, "Llama-2-70b-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-70b-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/Llama-2-70b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-70b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-70b-hf-q4f32_1-MLC"}, "Llama-2-7b-chat-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-7b-chat-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/Llama-2-7b-chat-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-7b-chat-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-7b-chat-hf-q4f32_1-MLC"}, "Llama-2-13b-chat-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-13b-chat-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/Llama-2-13b-chat-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-13b-chat-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-13b-chat-hf-q4f32_1-MLC"}, "Llama-2-70b-chat-hf": {"Model URL": "https://huggingface.co/meta-llama/Llama-2-70b-chat-hf", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/Llama-2-70b-chat-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama-2-70b-chat-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama-2-70b-chat-hf-q4f32_1-MLC"}, "CodeLlama-7b-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-7b-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/CodeLlama-7b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-7b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-7b-hf-q4f32_1-MLC"}, "CodeLlama-13b-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-13b-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/CodeLlama-13b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-13b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-13b-hf-q4f32_1-MLC"}, "CodeLlama-34b-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-34b-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/CodeLlama-34b-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-34b-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-34b-hf-q4f32_1-MLC"}, "CodeLlama-7b-Instruct-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-7b-Instruct-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-7b-Instruct-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-7b-Instruct-hf-q4f32_1-MLC"}, "CodeLlama-13b-Instruct-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-13b-Instruct-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-13b-Instruct-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-13b-Instruct-hf-q4f32_1-MLC"}, "CodeLlama-34b-Instruct-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-34b-Instruct-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-34b-Instruct-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-34b-Instruct-hf-q4f32_1-MLC"}, "CodeLlama-7b-Python-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-7b-Python-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-7b-Python-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-7b-Python-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-7b-Python-hf-q4f32_1-MLC"}, "CodeLlama-13b-Python-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-13b-Python-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-13b-Python-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-13b-Python-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-13b-Python-hf-q4f32_1-MLC"}, "CodeLlama-34b-Python-hf": {"Model URL": "https://huggingface.co/codellama/CodeLlama-34b-Python-hf", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "codellama_instruct", "q3f16_1": "HF://junrushao/CodeLlama-34b-Python-hf-q3f16_1-MLC", "q4f16_1": "HF://junrushao/CodeLlama-34b-Python-hf-q4f16_1-MLC", "q4f32_1": "HF://junrushao/CodeLlama-34b-Python-hf-q4f32_1-MLC"}, "Mistral-7B-Instruct-v0.2": {"Model URL": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2", "Model Type": "mistral", "Context Window": "32768", "Conversation Template": "mistral_default", "q3f16_1": "HF://junrushao/Mistral-7B-Instruct-v0.2-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Mistral-7B-Instruct-v0.2-q4f32_1-MLC"}, "NeuralHermes-2.5-Mistral-7B": {"Model URL": "https://huggingface.co/mlabonne/NeuralHermes-2.5-Mistral-7B", "Model Type": "mistral", "Context Window": "32768", "Conversation Template": "neural_hermes_mistral", "q3f16_1": "HF://junrushao/NeuralHermes-2.5-Mistral-7B-q3f16_1-MLC", "q4f16_1": "HF://junrushao/NeuralHermes-2.5-Mistral-7B-q4f16_1-MLC", "q4f32_1": "HF://junrushao/NeuralHermes-2.5-Mistral-7B-q4f32_1-MLC"}, "OpenHermes-2-Mistral-7B": {"Model URL": "https://huggingface.co/teknium/OpenHermes-2-Mistral-7B", "Model Type": "mistral", "Context Window": "32768", "Conversation Template": "open_hermes_mistral", "q3f16_1": "HF://junrushao/OpenHermes-2-Mistral-7B-q3f16_1-MLC", "q4f16_1": "HF://junrushao/OpenHermes-2-Mistral-7B-q4f16_1-MLC", "q4f32_1": "HF://junrushao/OpenHermes-2-Mistral-7B-q4f32_1-MLC"}, "WizardMath-7B-V1.1": {"Model URL": "https://huggingface.co/WizardLM/WizardMath-7B-V1.1", "Model Type": "mistral", "Context Window": "32768", "Conversation Template": "wizard_coder_or_math", "q3f16_1": "HF://junrushao/WizardMath-7B-V1.1-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardMath-7B-V1.1-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardMath-7B-V1.1-q4f32_1-MLC"}, "RedPajama-INCITE-Chat-3B-v1": {"Model URL": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Chat-3B-v1", "Model Type": "gpt_neox", "Context Window": "2048", "Conversation Template": "redpajama_chat", "q3f16_1": "HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q3f16_1-MLC", "q4f16_1": "HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC", "q4f32_1": "HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q4f32_1-MLC"}, "phi-2": {"Model URL": "https://huggingface.co/microsoft/phi-2", "Model Type": "phi-msft", "Context Window": "2048", "Conversation Template": "phi-2", "q3f16_1": "HF://junrushao/phi-2-q3f16_1-MLC", "q4f16_1": "HF://junrushao/phi-2-q4f16_1-MLC", "q4f32_1": "HF://junrushao/phi-2-q4f32_1-MLC"}, "phi-1_5": {"Model URL": "https://huggingface.co/microsoft/phi-1_5", "Model Type": "phi-msft", "Context Window": "2048", "Conversation Template": "phi-2", "q3f16_1": "HF://junrushao/phi-1_5-q3f16_1-MLC", "q4f16_1": "HF://junrushao/phi-1_5-q4f16_1-MLC", "q4f32_1": "HF://junrushao/phi-1_5-q4f32_1-MLC"}, "gpt2": {"Model URL": "https://huggingface.co/gpt2", "Model Type": "gpt2", "Context Window": "1024", "Conversation Template": "gpt2", "q3f16_1": "HF://junrushao/gpt2-q3f16_1-MLC", "q4f16_1": "HF://junrushao/gpt2-q4f16_1-MLC", "q4f32_1": "HF://junrushao/gpt2-q4f32_1-MLC"}, "gpt2-medium": {"Model URL": "https://huggingface.co/gpt2-medium", "Model Type": "gpt2", "Context Window": "1024", "Conversation Template": "gpt2", "q3f16_1": "HF://junrushao/gpt2-medium-q3f16_1-MLC", "q4f16_1": "HF://junrushao/gpt2-medium-q4f16_1-MLC", "q4f32_1": "HF://junrushao/gpt2-medium-q4f32_1-MLC"}, "vicuna-7b-v1.5": {"Model URL": "https://huggingface.co/lmsys/vicuna-7b-v1.5", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "vicuna_v1.1", "q3f16_1": "HF://junrushao/vicuna-7b-v1.5-q3f16_1-MLC", "q4f16_1": "HF://junrushao/vicuna-7b-v1.5-q4f16_1-MLC", "q4f32_1": "HF://junrushao/vicuna-7b-v1.5-q4f32_1-MLC"}, "WizardLM-7B-V1.0": {"Model URL": "https://huggingface.co/WizardLM/WizardLM-7B-V1.0", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "vicuna_v1.1", "q3f16_1": "HF://junrushao/WizardLM-7B-V1.0-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardLM-7B-V1.0-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardLM-7B-V1.0-q4f32_1-MLC"}, "WizardLM-13B-V1.1": {"Model URL": "https://huggingface.co/WizardLM/WizardLM-13B-V1.1", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "vicuna_v1.1", "q3f16_1": "HF://junrushao/WizardLM-13B-V1.1-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardLM-13B-V1.1-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardLM-13B-V1.1-q4f32_1-MLC"}, "WizardLM-30B-V1.0": {"Model URL": "https://huggingface.co/WizardLM/WizardLM-30B-V1.0", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "vicuna_v1.1", "q3f16_1": "HF://junrushao/WizardLM-30B-V1.0-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardLM-30B-V1.0-q3f16_1-MLC", "q4f32_1": "HF://junrushao/WizardLM-30B-V1.0-q4f32_1-MLC"}, "WizardLM-70B-V1.0": {"Model URL": "https://huggingface.co/WizardLM/WizardLM-70B-V1.0", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "vicuna_v1.1", "q3f16_1": "HF://junrushao/WizardLM-70B-V1.0-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardLM-70B-V1.0-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardLM-70B-V1.0-q4f32_1-MLC"}, "WizardCoder-15B-V1.0": {"Model URL": "https://huggingface.co/WizardLM/WizardCoder-15B-V1.0", "Model Type": "gpt_bigcode", "Context Window": "8192", "Conversation Template": "wizard_coder_or_math", "q3f16_1": "HF://junrushao/WizardCoder-15B-V1.0-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardCoder-15B-V1.0-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardCoder-15B-V1.0-q4f32_1-MLC"}, "OpenOrca-Platypus2-13B": {"Model URL": "https://huggingface.co/Open-Orca/OpenOrca-Platypus2-13B", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/OpenOrca-Platypus2-13B-q3f16_1-MLC", "q4f16_1": "HF://junrushao/OpenOrca-Platypus2-13B-q4f16_1-MLC", "q4f32_1": "HF://junrushao/OpenOrca-Platypus2-13B-q4f32_1-MLC"}, "Llama2-Chinese-7b-Chat": {"Model URL": "https://huggingface.co/FlagAlpha/Llama2-Chinese-7b-Chat", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/Llama2-Chinese-7b-Chat-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama2-Chinese-7b-Chat-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama2-Chinese-7b-Chat-q4f32_1-MLC"}, "Llama2-Chinese-13b-Chat": {"Model URL": "https://huggingface.co/FlagAlpha/Llama2-Chinese-13b-Chat", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/Llama2-Chinese-13b-Chat-q3f16_1-MLC", "q4f16_1": "HF://junrushao/Llama2-Chinese-13b-Chat-q4f16_1-MLC", "q4f32_1": "HF://junrushao/Llama2-Chinese-13b-Chat-q4f32_1-MLC"}, "llama2_7b_chat_uncensored": {"Model URL": "https://huggingface.co/georgesung/llama2_7b_chat_uncensored", "Model Type": "llama", "Context Window": "4096", "Conversation Template": "llama-default", "q3f16_1": "HF://junrushao/llama2_7b_chat_uncensored-q3f16_1-MLC", "q4f16_1": "HF://junrushao/llama2_7b_chat_uncensored-q4f16_1-MLC", "q4f32_1": "HF://junrushao/llama2_7b_chat_uncensored-q4f32_1-MLC"}, "WizardCoder-Python-34B-V1.0": {"Model URL": "https://huggingface.co/WizardLM/WizardCoder-Python-34B-V1.0", "Model Type": "llama", "Context Window": "16384", "Conversation Template": "wizard_coder_or_math", "q3f16_1": "HF://junrushao/WizardCoder-Python-34B-V1.0-q3f16_1-MLC", "q4f16_1": "HF://junrushao/WizardCoder-Python-34B-V1.0-q4f16_1-MLC", "q4f32_1": "HF://junrushao/WizardCoder-Python-34B-V1.0-q4f32_1-MLC"}, "open_llama_3b": {"Model URL": "https://huggingface.co/openlm-research/open_llama_3b", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "llama-default", "q3f16_1": "HF://junrushao/open_llama_3b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/open_llama_3b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/open_llama_3b-q4f32_1-MLC"}, "open_llama_7b": {"Model URL": "https://huggingface.co/openlm-research/open_llama_7b", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "llama-default", "q3f16_1": "HF://junrushao/open_llama_7b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/open_llama_7b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/open_llama_7b-q4f32_1-MLC"}, "open_llama_13b": {"Model URL": "https://huggingface.co/openlm-research/open_llama_13b", "Model Type": "llama", "Context Window": "2048", "Conversation Template": "llama-default", "q3f16_1": "HF://junrushao/open_llama_13b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/open_llama_13b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/open_llama_13b-q4f32_1-MLC"}, "YuLan-Chat-2-13b-fp16": {"Model URL": "https://huggingface.co/yulan-team/YuLan-Chat-2-13b-fp16", "Model Type": "llama", "Context Window": "8192", "Conversation Template": "llama-2", "q3f16_1": "HF://junrushao/YuLan-Chat-2-13b-fp16-q3f16_1-MLC", "q4f16_1": "HF://junrushao/YuLan-Chat-2-13b-fp16-q4f16_1-MLC", "q4f32_1": "HF://junrushao/YuLan-Chat-2-13b-fp16-q4f32_1-MLC"}, "dolly-v2-12b": {"Model URL": "https://huggingface.co/databricks/dolly-v2-12b", "Model Type": "gpt_neox", "Context Window": "2048", "Conversation Template": "dolly", "q3f16_1": "HF://junrushao/dolly-v2-12b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/dolly-v2-12b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/dolly-v2-12b-q4f32_1-MLC"}, "pythia-1.4b": {"Model URL": "https://huggingface.co/EleutherAI/pythia-1.4b", "Model Type": "gpt_neox", "Context Window": "2048", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/pythia-1.4b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/pythia-1.4b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/pythia-1.4b-q4f32_1-MLC"}, "stablecode-instruct-alpha-3b": {"Model URL": "https://huggingface.co/stabilityai/stablecode-instruct-alpha-3b", "Model Type": "gpt_neox", "Context Window": "4096", "Conversation Template": "stablecode_instruct", "q3f16_1": "HF://junrushao/stablecode-instruct-alpha-3b-q3f16_1-MLC", "q4f16_1": "HF://junrushao/stablecode-instruct-alpha-3b-q4f16_1-MLC", "q4f32_1": "HF://junrushao/stablecode-instruct-alpha-3b-q4f32_1-MLC"}, "starcoder": {"Model URL": "https://huggingface.co/bigcode/starcoder", "Model Type": "gpt_bigcode", "Context Window": "8192", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/starcoder-q3f16_1-MLC", "q4f16_1": "HF://junrushao/starcoder-q4f16_1-MLC", "q4f32_1": "HF://junrushao/starcoder-q4f32_1-MLC"}, "gpt_bigcode-santacoder": {"Model URL": "https://huggingface.co/bigcode/gpt_bigcode-santacoder", "Model Type": "gpt_bigcode", "Context Window": "2048", "Conversation Template": "LM", "q3f16_1": "HF://junrushao/gpt_bigcode-santacoder-q3f16_1-MLC", "q4f16_1": "HF://junrushao/gpt_bigcode-santacoder-q4f16_1-MLC", "q4f32_1": "HF://junrushao/gpt_bigcode-santacoder-q4f32_1-MLC"}}
\ No newline at end of file
diff --git a/wasted/phi_model_old.py b/wasted/phi_model_old.py
new file mode 100644
index 0000000000..8e520f6653
--- /dev/null
+++ b/wasted/phi_model_old.py
@@ -0,0 +1,451 @@
+"""
+Implementation for Phi architecture.
+TODO: add docstring
+"""
+import dataclasses
+from typing import Any, Dict, Optional, Union
+
+from tvm import te, tir
+from tvm.relax.frontend import nn
+from tvm.relax.frontend.nn import Tensor, op
+
+from mlc_chat import op as op_ext
+from mlc_chat.support import logging
+from mlc_chat.support import tensor_parallel as tp
+from mlc_chat.support.config import ConfigBase
+from mlc_chat.support.style import bold
+
+logger = logging.getLogger(__name__)
+
+
+@dataclasses.dataclass
+class Phi1Config(ConfigBase):  # pylint: disable=too-many-instance-attributes
+    """Configuration of the Phi-1/Phi-1.5 model."""
+
+    vocab_size: int = 51200
+    hidden_size: int = 2048
+    intermediate_size: int = 8192
+    num_hidden_layers: int = 24
+    num_attention_heads: int = 32
+    layer_norm_eps: float = 1e-5
+    position_embedding_base: int = 0
+    partial_rotary_factor: float = 0.5
+    num_key_value_heads: int = 0
+    context_window_size: int = 0
+    prefill_chunk_size: int = 0
+    head_dim: int = 0
+    tensor_parallel_shards: int = 1
+    kwargs: Dict[str, Any] = dataclasses.field(default_factory=dict)
+
+    def __post_init__(self):
+        if self.position_embedding_base == 0:
+            if "rope_theta" in self.kwargs:
+                self.position_embedding_base = self.kwargs.pop("rope_theta")
+            else:
+                self.position_embedding_base = 10000
+        if self.context_window_size == 0:
+            for name in ["max_position_embeddings", "max_sequence_length"]:
+                if name in self.kwargs:
+                    self.context_window_size = self.kwargs.pop(name)
+                    logger.info(
+                        "%s not found in config.json. Falling back to %s (%d)",
+                        bold("context_window_size"),
+                        bold(name),
+                        self.context_window_size,
+                    )
+                    break
+            else:
+                raise ValueError(
+                    "Unable to determine the maxmimum sequence length, because none of "
+                    "`context_window_size`, `max_position_embeddings` or `max_sequence_length` is "
+                    "provided in `config.json`."
+                )
+        if self.prefill_chunk_size == 0:
+            self.prefill_chunk_size = self.context_window_size
+        if self.prefill_chunk_size > self.context_window_size:
+            self.prefill_chunk_size = self.context_window_size
+        if self.num_key_value_heads == 0 or self.num_key_value_heads is None:
+            self.num_key_value_heads = self.num_attention_heads
+        if self.intermediate_size == 0 or self.intermediate_size is None:
+            self.intermediate_size = 4 * self.hidden_size
+        if self.head_dim == 0:
+            self.head_dim = self.hidden_size // self.num_attention_heads
+        assert self.head_dim * self.num_attention_heads == self.hidden_size
+        assert self.num_attention_heads % self.num_key_value_heads == 0
+
+
+@dataclasses.dataclass
+class PhiConfig(ConfigBase):  # pylint: disable=too-many-instance-attributes
+    """Configuration of the Phi-2 model."""
+
+    model_type: str  # "phi", "phi-msft", "mixformer-sequential"
+    vocab_size: int = 51200
+    n_positions: int = 2048
+    n_embd: int = 2560
+    n_layer: int = 32
+    n_inner: int = 0
+    n_head: int = 32
+    rotary_dim: int = 32
+    position_embedding_base: int = 0
+    layer_norm_epsilon: float = 1e-5
+    context_window_size: int = 0
+    prefill_chunk_size: int = 0
+    n_head_kv: int = 0
+    head_dim: int = 0
+    tensor_parallel_shards: int = 1
+    kwargs: Dict[str, Any] = dataclasses.field(default_factory=dict)
+
+    def __post_init__(self):
+        if self.position_embedding_base == 0:
+            if "rope_theta" in self.kwargs:
+                self.position_embedding_base = self.kwargs.pop("rope_theta")
+            else:
+                self.position_embedding_base = 10000
+        if self.context_window_size == 0:
+            for name in ["max_position_embeddings", "max_sequence_length"]:
+                if name in self.kwargs:
+                    self.context_window_size = self.kwargs.pop(name)
+                    logger.info(
+                        "%s not found in config.json. Falling back to %s (%d)",
+                        bold("context_window_size"),
+                        bold(name),
+                        self.context_window_size,
+                    )
+                    break
+            else:
+                self.context_window_size = self.n_positions
+                logger.info(
+                    "%s not found in config.json. Falling back to %s (%d)",
+                    bold("context_window_size"),
+                    "n_positions",
+                    self.context_window_size,
+                )
+        if self.prefill_chunk_size == 0:
+            self.prefill_chunk_size = self.context_window_size
+        if self.prefill_chunk_size > self.context_window_size:
+            self.prefill_chunk_size = self.context_window_size
+        if self.n_head_kv == 0 or self.n_head_kv is None:
+            self.n_head_kv = self.n_head
+        if self.n_inner == 0 or self.n_inner is None:
+            self.n_inner = 4 * self.n_embd
+        if self.head_dim == 0:
+            self.head_dim = self.n_embd // self.n_head
+        assert self.head_dim * self.n_head == self.n_embd
+        assert self.n_head % self.n_head_kv == 0
+
+    @staticmethod
+    def from_phi1(config: Phi1Config) -> "PhiConfig":
+        "Build PhiConig from a Phi1Config."
+        return PhiConfig(
+            model_type="phi",
+            vocab_size=config.vocab_size,
+            n_positions=config.context_window_size,
+            n_embd=config.hidden_size,
+            n_layer=config.num_hidden_layers,
+            n_inner=config.intermediate_size,
+            n_head=config.num_attention_heads,
+            rotary_dim=int(config.partial_rotary_factor * config.head_dim),
+            position_embedding_base=config.position_embedding_base,
+            layer_norm_epsilon=config.layer_norm_eps,
+            context_window_size=config.context_window_size,
+            prefill_chunk_size=config.prefill_chunk_size,
+            n_head_kv=config.num_key_value_heads,
+            head_dim=config.head_dim,
+            tensor_parallel_shards=config.tensor_parallel_shards,
+            kwargs=config.kwargs,
+        )
+
+
+# pylint: disable=invalid-name,missing-docstring
+
+
+class PhiMLP(nn.Module):
+    def __init__(self, config: PhiConfig):
+        super().__init__()
+        self.intermediate_size = config.n_inner // config.tensor_parallel_shards
+        self.fc1 = nn.Linear(config.n_embd, self.intermediate_size)
+        self.fc2 = nn.Linear(self.intermediate_size, config.n_embd)
+
+    def forward(self, hidden_states: Tensor):
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = op.gelu(hidden_states, approximate="tanh")
+        hidden_states = self.fc2(hidden_states)
+
+        return hidden_states
+
+
+class PhiCrossAttention(nn.Module):
+    def __init__(self, config: PhiConfig):  # pylint: disable=unused-argument
+        super().__init__()
+
+    def forward(self, q: Tensor, k: Tensor, v: Tensor, attention_mask: Tensor):
+        output = op_ext.attention(q, k, v, casual_mask=attention_mask, qk_dtype="float32")
+        return output
+
+
+class PhiMHA(nn.Module):  # pylint: disable=too-many-instance-attributes
+    def __init__(self, config: PhiConfig):
+        self.rope_theta = config.position_embedding_base
+        self.rotary_dim = config.rotary_dim
+        self.n_head = config.n_head // config.tensor_parallel_shards
+        assert (
+            config.n_head % config.tensor_parallel_shards == 0
+        ), f"n_head({config.n_head}) must be divisible by tensor_parallel_shards"
+        self.n_head_kv = config.n_head_kv // config.tensor_parallel_shards
+        assert (
+            config.n_head_kv % config.tensor_parallel_shards == 0
+        ), f"n_head({config.n_head_kv}) must be divisible by tensor_parallel_shards"
+        self.head_dim = config.head_dim
+        op_size = self.head_dim * (self.n_head + 2 * self.n_head_kv)
+        hidden_size = config.n_embd
+
+        self.Wqkv = nn.Linear(hidden_size, op_size, bias=True)
+        self.out_proj = nn.Linear(self.n_head * self.head_dim, hidden_size, bias=True)
+        self.inner_cross_attn = PhiCrossAttention(config)
+        self.k_cache = nn.KVCache(config.context_window_size, [self.n_head_kv, self.head_dim])
+        self.v_cache = nn.KVCache(config.context_window_size, [self.n_head_kv, self.head_dim])
+
+    def forward(self, x: Tensor, attention_mask: Tensor, total_seq_len: tir.Var):
+        d, h_q, h_kv, t = self.head_dim, self.n_head, self.n_head_kv, total_seq_len
+        b, s, _ = x.shape
+        assert b == 1, "Only support batch size 1 at this moment."
+        # Step 1. QKV Projection
+        qkv = self.Wqkv(x)
+        # if self.idx == 0:
+        #     op.print_inter(qkv, self.tensor_parallel_shards)
+        qkv = op.reshape(qkv, (b, s, h_q + h_kv + h_kv, d))
+        # Step 2. Apply QK rotary embedding
+        q, k, v = op_ext.llama_rope(qkv, t, self.rope_theta, h_q, h_kv, rotary_dim=self.rotary_dim)
+        # if self.idx == 0:
+        #     op.print_inter(q, self.tensor_parallel_shards)
+        # Step 3. Query and update KVCache
+        self.k_cache.append(op.squeeze(k, axis=0))
+        self.v_cache.append(op.squeeze(v, axis=0))
+        k = self.k_cache.view(t)
+        v = self.v_cache.view(t)
+        # Step 4. Compute softmax(Q @ K^T / sqrt(d)) @ V
+        # if self.idx == 0:
+        #     op.print_inter(k, self.tensor_parallel_shards)
+        output = self.inner_cross_attn(q, k, v, attention_mask)
+        # if self.idx == 0:
+        #     op.print_inter(output, self.tensor_parallel_shards)
+        # Step 5. Apply output projection
+        self.out_proj.idx = self.idx
+        return self.out_proj(output)
+
+
+class PhiParallelBlock(nn.Module):
+    def __init__(self, config: PhiConfig):
+        super().__init__()
+
+        self.ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.mixer = PhiMHA(config)
+        self.mlp = PhiMLP(config)
+
+        def _set_tp():
+            def _set(param, hint):
+                param.attrs["shard_strategy"] = hint
+
+            hd = config.head_dim
+            q = self.mixer.n_head * hd
+            k = self.mixer.n_head_kv * hd
+            v = self.mixer.n_head_kv * hd
+            _set(
+                self.mixer.Wqkv.weight,
+                tp.ShardSingleDim("_shard_qkv_weight", segs=[q, k, v], dim=0),
+            )
+            _set(self.mixer.Wqkv.bias, tp.ShardSingleDim("_shard_qkv_bias", segs=[q, k, v], dim=0))
+            _set(self.mixer.out_proj.weight, tp.ShardSingleDim("_shard_o_weight", dim=1))
+            _set(self.mlp.fc1.weight, tp.ShardSingleDim("_shard_mlp_fc1_weight", dim=0))
+            _set(self.mlp.fc1.bias, tp.ShardSingleDim("_shard_mlp_fc1_bias", dim=0))
+            _set(self.mlp.fc2.weight, tp.ShardSingleDim("_shard_mlp_fc2_weight", dim=1))
+
+        self.tensor_parallel_shards = config.tensor_parallel_shards
+        _set_tp()
+
+    def forward(self, hidden_states: Tensor, attention_mask: Tensor, total_seq_len: tir.Var):
+        residual = hidden_states
+        hidden_states = self.ln(hidden_states)
+
+        self.mixer.idx = self.idx
+        self.mlp.idx = self.idx
+        self.mixer.tensor_parallel_shards = self.tensor_parallel_shards
+        self.mlp.tensor_parallel_shards = self.tensor_parallel_shards
+
+        # attn_outputs = (
+        #     self.mixer(
+        #         hidden_states,
+        #         attention_mask,
+        #         total_seq_len,
+        #     )
+        #     - self.mixer.out_proj.bias
+        # )
+
+        # if self.idx == 0:
+        #     op.print_inter(attn_outputs, self.tensor_parallel_shards)
+
+        # feed_forward_hidden_states = self.mlp(hidden_states) - self.mlp.fc2.bias
+
+        # if self.idx == 0:
+        #     op.print_inter(feed_forward_hidden_states, self.tensor_parallel_shards)
+
+        with tp.shard_bias(self.mixer.out_proj, self.tensor_parallel_shards), tp.shard_bias(
+            self.mlp.fc2, self.tensor_parallel_shards
+        ):
+            attn_outputs = self.mixer(
+                hidden_states,
+                attention_mask,
+                total_seq_len,
+            )
+
+            # if self.idx == 0:
+            #     op.print_inter(attn_outputs, self.tensor_parallel_shards)
+
+            feed_forward_hidden_states = self.mlp(hidden_states)
+
+            # if self.idx == 0:
+            #     op.print_inter(feed_forward_hidden_states, self.tensor_parallel_shards)
+
+        def _apply_parallel_residual(attn_out, mlp_out, residual):
+            if self.tensor_parallel_shards > 1:
+                return op.ccl_allreduce(
+                    attn_out + mlp_out + residual / self.tensor_parallel_shards, "sum"
+                )
+            return attn_out + mlp_out + residual
+
+        hidden_states = _apply_parallel_residual(attn_outputs, feed_forward_hidden_states, residual)
+
+        # if self.tensor_parallel_shards > 1:
+        # hidden_states = hidden_states + self.mixer.out_proj.bias + self.mlp.fc2.bias
+
+        # if self.idx == 0:
+        #     op.print_inter(hidden_states, self.tensor_parallel_shards)
+
+        return hidden_states
+
+
+class PhiCausalLMHead(nn.Module):
+    def __init__(self, config: PhiConfig) -> None:
+        super().__init__()
+
+        self.ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.linear = nn.Linear(config.n_embd, "vocab_size")
+
+    def forward(self, hidden_states: Tensor):
+        hidden_states = self.ln(hidden_states)
+        logits = self.linear(hidden_states)
+
+        if logits.dtype != "float32":
+            logits = logits.astype("float32")
+        return logits
+
+
+class PhiModel(nn.Module):
+    def __init__(self, config: PhiConfig) -> None:
+        super().__init__()
+        self.embd = nn.Embedding("vocab_size", config.n_embd)
+        self.h = nn.ModuleList([PhiParallelBlock(config) for i in range(config.n_layer)])
+        self.tensor_parallel_shards = config.tensor_parallel_shards
+
+    def forward(self, input_ids: Tensor, total_seq_len: tir.Var, attention_mask: Tensor):
+        if self.tensor_parallel_shards > 1:
+            input_ids = op.ccl_broadcast_from_worker0(input_ids)
+        hidden_states = self.embd(input_ids)
+        # op.print_inter(hidden_states)
+        for i, layer in enumerate(self.h):
+            layer.idx = i
+            hidden_states = layer(hidden_states, attention_mask, total_seq_len)
+            # if i == 0:
+            #     op.print_inter(hidden_states)
+
+        return hidden_states
+
+
+class PhiForCausalLM(nn.Module):
+    def __init__(self, config: Union[PhiConfig, Phi1Config]) -> None:
+        super().__init__()
+
+        if isinstance(config, Phi1Config):
+            config = PhiConfig.from_phi1(config)
+
+        self.transformer = PhiModel(config)
+        self.lm_head = PhiCausalLMHead(config)
+        self.dtype = "float32"
+
+    def to(self, dtype: Optional[str] = None):
+        super().to(dtype=dtype)
+        if dtype is not None:
+            self.dtype = dtype
+
+    def forward(self, input_ids: Tensor, total_seq_len: tir.Var, attention_mask: Tensor):
+        def _index(x: te.Tensor):  # x[:-1,:]
+            b, s, d = x.shape
+            return te.compute((b, 1, d), lambda i, _, k: x[i, s - 1, k], name="index")
+
+        hidden_states = self.transformer(input_ids, total_seq_len, attention_mask)
+        hidden_states = op.tensor_expr_op(_index, name_hint="index", args=[hidden_states])
+        lm_logits = self.lm_head(hidden_states)
+
+        return lm_logits
+
+    def prefill(self, inputs: Tensor, total_seq_len: tir.Var):
+        def _attention_mask(batch_size, seq_len, total_seq_len):
+            return te.compute(
+                (batch_size, 1, seq_len, total_seq_len),
+                lambda b, _, i, j: tir.if_then_else(
+                    i < j - (total_seq_len - seq_len),
+                    tir.min_value(self.dtype),
+                    tir.max_value(self.dtype),
+                ),
+                name="attention_mask_prefill",
+            )
+
+        batch_size, seq_len = inputs.shape
+        attention_mask = op.tensor_expr_op(
+            _attention_mask,
+            name_hint="attention_mask_prefill",
+            args=[batch_size, seq_len, total_seq_len],
+        )
+        return self.forward(inputs, total_seq_len, attention_mask)
+
+    def decode(self, inputs: Tensor, total_seq_len: tir.Var):
+        batch_size, seq_len = inputs.shape
+        attention_mask = op.full(
+            shape=[batch_size, 1, seq_len, total_seq_len],
+            fill_value=tir.max_value(self.dtype),
+            dtype=self.dtype,
+        )
+        return self.forward(inputs, total_seq_len, attention_mask)
+
+    def softmax_with_temperature(self, logits: Tensor, temperature: Tensor):
+        return op.softmax(logits / temperature, axis=-1)
+
+    def get_default_spec(self):
+        batch_size = 1
+        mod_spec = {
+            "prefill": {
+                "inputs": nn.spec.Tensor([batch_size, "seq_len"], "int32"),
+                "total_seq_len": int,
+                "$": {
+                    "param_mode": "packed",
+                    "effect_mode": "packed",
+                },
+            },
+            "decode": {
+                "inputs": nn.spec.Tensor([batch_size, 1], "int32"),
+                "total_seq_len": int,
+                "$": {
+                    "param_mode": "packed",
+                    "effect_mode": "packed",
+                },
+            },
+            "softmax_with_temperature": {
+                "logits": nn.spec.Tensor([1, 1, "vocab_size"], "float32"),
+                "temperature": nn.spec.Tensor([], "float32"),
+                "$": {
+                    "param_mode": "none",
+                    "effect_mode": "none",
+                },
+            },
+        }
+        return nn.spec.ModuleSpec.from_raw(mod_spec, self)
diff --git a/wasted/sample_mlc_chat.py b/wasted/sample_mlc_chat.py
new file mode 100644
index 0000000000..b97c700b90
--- /dev/null
+++ b/wasted/sample_mlc_chat.py
@@ -0,0 +1,27 @@
+from mlc_chat import ChatConfig, ChatModule
+from mlc_chat.callback import StreamToStdout
+
+# From the mlc-llm directory, run
+# $ python examples/python/sample_mlc_chat.py
+
+# Create a ChatModule instance
+# Multiple GPU
+cm = ChatModule(
+    # model="dist/llama-2-7b-chat-hf-q4f16_1/params",
+    model="dist/llama-2-7b-chat-hf-q4f16_1-presharded-2gpu/params",
+    chat_config=ChatConfig(tensor_parallel_shards=2)
+    # model="dist/Mixtral-8x7B-Instruct-v0.1-q0f16/params",
+    # chat_config=ChatConfig(tensor_parallel_shards=2),
+)
+
+# Generate a response for a given prompt
+output = cm.generate(
+    prompt="What is the meaning of life?",
+    progress_callback=StreamToStdout(callback_interval=2),
+)
+
+# Print prefill and decode performance statistics
+print(f"Statistics: {cm.stats()}\n")
+
+# Reset the chat module by
+cm.reset_chat()
diff --git a/wasted/sheet.csv b/wasted/sheet.csv
new file mode 100644
index 0000000000..be037c7d47
--- /dev/null
+++ b/wasted/sheet.csv
@@ -0,0 +1,29 @@
+Name (last piece of Model URL),Model URL,Model Type,Context Window,Conversation Template,q3f16_1,q4f16_1,q4f32_1
+Llama-2-7b-hf,https://huggingface.co/meta-llama/Llama-2-7b-hf,llama,4096,LM,HF://junrushao/Llama-2-7b-hf-q3f16_1-MLC,HF://junrushao/Llama-2-7b-hf-q4f16_1-MLC,HF://junrushao/Llama-2-7b-hf-q4f32_1-MLC
+Llama-2-13b-hf,https://huggingface.co/meta-llama/Llama-2-13b-hf,llama,4096,LM,HF://junrushao/Llama-2-13b-hf-q3f16_1-MLC,HF://junrushao/Llama-2-13b-hf-q4f16_1-MLC,HF://junrushao/Llama-2-13b-hf-q4f32_1-MLC
+Llama-2-70b-hf,https://huggingface.co/meta-llama/Llama-2-70b-hf,llama,4096,LM,HF://junrushao/Llama-2-70b-hf-q3f16_1-MLC,HF://junrushao/Llama-2-70b-hf-q4f16_1-MLC,HF://junrushao/Llama-2-70b-hf-q4f32_1-MLC
+Llama-2-7b-chat-hf,https://huggingface.co/meta-llama/Llama-2-7b-chat-hf,llama,4096,llama-2,HF://junrushao/Llama-2-7b-chat-hf-q3f16_1-MLC,HF://junrushao/Llama-2-7b-chat-hf-q4f16_1-MLC,HF://junrushao/Llama-2-7b-chat-hf-q4f32_1-MLC
+Llama-2-13b-chat-hf,https://huggingface.co/meta-llama/Llama-2-13b-chat-hf,llama,4096,llama-2,HF://junrushao/Llama-2-13b-chat-hf-q3f16_1-MLC,HF://junrushao/Llama-2-13b-chat-hf-q4f16_1-MLC,HF://junrushao/Llama-2-13b-chat-hf-q4f32_1-MLC
+Llama-2-70b-chat-hf,https://huggingface.co/meta-llama/Llama-2-70b-chat-hf,llama,4096,llama-2,HF://junrushao/Llama-2-70b-chat-hf-q3f16_1-MLC,HF://junrushao/Llama-2-70b-chat-hf-q4f16_1-MLC,HF://junrushao/Llama-2-70b-chat-hf-q4f32_1-MLC
+CodeLlama-7b-hf,https://huggingface.co/codellama/CodeLlama-7b-hf,llama,16384,LM,HF://junrushao/CodeLlama-7b-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-7b-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-7b-hf-q4f32_1-MLC
+CodeLlama-13b-hf,https://huggingface.co/codellama/CodeLlama-13b-hf,llama,16384,LM,HF://junrushao/CodeLlama-13b-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-13b-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-13b-hf-q4f32_1-MLC
+CodeLlama-34b-hf,https://huggingface.co/codellama/CodeLlama-34b-hf,llama,16384,LM,HF://junrushao/CodeLlama-34b-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-34b-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-34b-hf-q4f32_1-MLC
+CodeLlama-7b-Instruct-hf,https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-7b-Instruct-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-7b-Instruct-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-7b-Instruct-hf-q4f32_1-MLC
+CodeLlama-13b-Instruct-hf,https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-13b-Instruct-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-13b-Instruct-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-13b-Instruct-hf-q4f32_1-MLC
+CodeLlama-34b-Instruct-hf,https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-34b-Instruct-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-34b-Instruct-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-34b-Instruct-hf-q4f32_1-MLC
+CodeLlama-7b-Python-hf,https://huggingface.co/codellama/CodeLlama-7b-Python-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-7b-Python-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-7b-Python-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-7b-Python-hf-q4f32_1-MLC
+CodeLlama-13b-Python-hf,https://huggingface.co/codellama/CodeLlama-13b-Python-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-13b-Python-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-13b-Python-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-13b-Python-hf-q4f32_1-MLC
+CodeLlama-34b-Python-hf,https://huggingface.co/codellama/CodeLlama-34b-Python-hf,llama,16384,codellama_instruct,HF://junrushao/CodeLlama-34b-Python-hf-q3f16_1-MLC,HF://junrushao/CodeLlama-34b-Python-hf-q4f16_1-MLC,HF://junrushao/CodeLlama-34b-Python-hf-q4f32_1-MLC
+Mistral-7B-Instruct-v0.2,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2,mistral,32768,mistral_default,HF://junrushao/Mistral-7B-Instruct-v0.2-q3f16_1-MLC,HF://junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC,HF://junrushao/Mistral-7B-Instruct-v0.2-q4f32_1-MLC
+NeuralHermes-2.5-Mistral-7B,https://huggingface.co/mlabonne/NeuralHermes-2.5-Mistral-7B,mistral,32768,neural_hermes_mistral,HF://junrushao/NeuralHermes-2.5-Mistral-7B-q3f16_1-MLC,HF://junrushao/NeuralHermes-2.5-Mistral-7B-q4f16_1-MLC,HF://junrushao/NeuralHermes-2.5-Mistral-7B-q4f32_1-MLC
+OpenHermes-2-Mistral-7B,https://huggingface.co/teknium/OpenHermes-2-Mistral-7B,mistral,32768,open_hermes_mistral,HF://junrushao/OpenHermes-2-Mistral-7B-q3f16_1-MLC,HF://junrushao/OpenHermes-2-Mistral-7B-q4f16_1-MLC,HF://junrushao/OpenHermes-2-Mistral-7B-q4f32_1-MLC
+RedPajama-INCITE-Chat-3B-v1,https://huggingface.co/togethercomputer/RedPajama-INCITE-Chat-3B-v1,gpt_neox,2048,redpajama_chat,HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q3f16_1-MLC,HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC,HF://junrushao/RedPajama-INCITE-Chat-3B-v1-q4f32_1-MLC
+phi-2,https://huggingface.co/microsoft/phi-2,phi-msft,2048,phi-2,HF://junrushao/phi-2-q3f16_1-MLC,HF://junrushao/phi-2-q4f16_1-MLC,HF://junrushao/phi-2-q4f32_1-MLC
+phi-1_5,https://huggingface.co/microsoft/phi-1_5,phi-msft,2048,phi-2,HF://junrushao/phi-1_5-q3f16_1-MLC,HF://junrushao/phi-1_5-q4f16_1-MLC,HF://junrushao/phi-1_5-q4f32_1-MLC
+vicuna-7b-v1.5,https://huggingface.co/lmsys/vicuna-7b-v1.5,llama,4096,vicuna_v1.1,HF://junrushao/vicuna-7b-v1.5-q3f16_1-MLC,HF://junrushao/vicuna-7b-v1.5-q4f16_1-MLC,HF://junrushao/vicuna-7b-v1.5-q4f32_1-MLC
+gpt2,https://huggingface.co/gpt2,gpt2,1024,gpt2,HF://junrushao/gpt2-q3f16_1-MLC,HF://junrushao/gpt2-q4f16_1-MLC,HF://junrushao/gpt2-q4f32_1-MLC
+gpt2-medium,https://huggingface.co/gpt2-medium,gpt2,1024,gpt2,HF://junrushao/gpt2-medium-q3f16_1-MLC,HF://junrushao/gpt2-medium-q4f16_1-MLC,HF://junrushao/gpt2-medium-q4f32_1-MLC
+OpenOrca-Platypus2-13B,https://huggingface.co/Open-Orca/OpenOrca-Platypus2-13B,llama,4096,llama-2,HF://junrushao/OpenOrca-Platypus2-13B-q3f16_1-MLC,HF://junrushao/OpenOrca-Platypus2-13B-q4f16_1-MLC,HF://junrushao/OpenOrca-Platypus2-13B-q4f32_1-MLC
+Llama2-Chinese-7b-Chat,https://huggingface.co/FlagAlpha/Llama2-Chinese-7b-Chat,llama,4096,llama-2,HF://junrushao/Llama2-Chinese-7b-Chat-q3f16_1-MLC,HF://junrushao/Llama2-Chinese-7b-Chat-q4f16_1-MLC,HF://junrushao/Llama2-Chinese-7b-Chat-q4f32_1-MLC
+Llama2-Chinese-13b-Chat,https://huggingface.co/FlagAlpha/Llama2-Chinese-13b-Chat,llama,4096,llama-2,HF://junrushao/Llama2-Chinese-13b-Chat-q3f16_1-MLC,HF://junrushao/Llama2-Chinese-13b-Chat-q4f16_1-MLC,HF://junrushao/Llama2-Chinese-13b-Chat-q4f32_1-MLC
+YuLan-Chat-2-13b-fp16,https://huggingface.co/yulan-team/YuLan-Chat-2-13b-fp16,llama,8192,llama-2,HF://junrushao/YuLan-Chat-2-13b-fp16-q3f16_1-MLC,HF://junrushao/YuLan-Chat-2-13b-fp16-q4f16_1-MLC,HF://junrushao/YuLan-Chat-2-13b-fp16-q4f32_1-MLC
\ No newline at end of file
diff --git a/wasted/test000.py b/wasted/test000.py
new file mode 100644
index 0000000000..4333be36ad
--- /dev/null
+++ b/wasted/test000.py
@@ -0,0 +1,54 @@
+# pylint: skip-file
+import argparse
+
+import numpy as np
+import torch
+import tvm
+from transformers import AutoTokenizer
+from tvm import relax
+from tvm.relax.frontend.nn import IOEffect
+from tvm.runtime import disco as di
+
+import mlc_chat
+
+
+def deploy_to_pipeline(args) -> None:
+    devices = list(range(args.tensor_parallel_shards))
+    print(devices)
+    sess = di.ProcessSession(
+        num_workers=args.tensor_parallel_shards, entrypoint="mlc_chat.cli.worker"
+    )
+    sess.init_ccl("nccl", *devices)
+    print(args.model_lib)
+    mod = sess.load_vm_module(args.model_lib)
+    create_kv_cache = mod["create_kv_cache"]
+
+    loader = sess.get_global_func("mlc.loader.LoadMultiGPU")
+    params = loader(args.params, mod)
+    exit(0)
+
+
+def _parse_args():
+    args = argparse.ArgumentParser()
+    args.add_argument("--model-lib", type=str, required=True)
+    args.add_argument("--params", type=str, required=True)
+    args.add_argument("--tensor-parallel-shards", type=int, default=4)
+    args.add_argument("--primary-device", type=str, default="auto")
+    args.add_argument("--prompt", type=str, default="What is the meaning of life?")
+    args.add_argument("--time-eval", default=False, action="store_true")
+    args.add_argument("--skip-rounds", type=int, default=0)
+    parsed = args.parse_args()
+
+    if parsed.primary_device == "auto":
+        if tvm.cuda().exist:
+            parsed.primary_device = "cuda"
+        elif tvm.metal().exist:
+            parsed.primary_device = "metal"
+        else:
+            raise ValueError("Cannot auto deduce device-name, please set it")
+    return parsed
+
+
+if __name__ == "__main__":
+    args = _parse_args()
+    deploy_to_pipeline(args)
diff --git a/wasted/test_7mistral.sh b/wasted/test_7mistral.sh
new file mode 100755
index 0000000000..648abb5e11
--- /dev/null
+++ b/wasted/test_7mistral.sh
@@ -0,0 +1,11 @@
+NUM_SHARDS=2
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+OUTPUT_PATH=./dist/Mixtral-8x7B-Instruct-v0.1-q0f16/params
+
+# python -m mlc_chat gen_config --model $MODEL_PATH/config.json --quantization q4f16_1 --conv-template llama-2 --context-window-size 4096 --num-shards $NUM_SHARDS -o $OUTPUT_PATH
+
+# python -m mlc_chat compile --model $OUTPUT_PATH -o $OUTPUT_PATH/llama.so
+
+# python -m mlc_chat convert_weight --model $MODEL_PATH --quantization q4f16_1 -o $OUTPUT_PATH
+
+CUDA_VISIBLE_DEVICES=6,7 python -m mlc_chat.cli.benchmark --model ./dist/Mixtral-8x7B-Instruct-v0.1-q0f16/params --model-lib ./dist/Mixtral-8x7B-Instruct-v0.1-q0f16/Mixtral-8x7B-Instruct-v0.1-q0f16-cuda.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256 --num-shards 2
diff --git a/wasted/test_awq.sh b/wasted/test_awq.sh
new file mode 100755
index 0000000000..9742bb5151
--- /dev/null
+++ b/wasted/test_awq.sh
@@ -0,0 +1,10 @@
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+OUTPUT_PATH=./dist/new-llama-awq-mit/
+
+# python -m mlc_chat compile --model llama2_7b --quantization q4f16_awq -o $OUTPUT_PATH/llama.so --context-window-size 4096
+
+python -m mlc_chat convert_weight $MODEL_PATH --quantization q4f16_awq -o $OUTPUT_PATH --source-format awq --source dist/models/llama-2-7b-chat-w4-g128-awq.pt
+
+# python -m mlc_chat gen_mlc_chat_config --model $MODEL_PATH/config.json --quantization q4f16_awq -o $OUTPUT_PATH --conv-template llama-2
+
+# CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH --model-lib $OUTPUT_PATH/llama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/wasted/test_awq_gemm.sh b/wasted/test_awq_gemm.sh
new file mode 100755
index 0000000000..75176becf5
--- /dev/null
+++ b/wasted/test_awq_gemm.sh
@@ -0,0 +1,14 @@
+#!/bin/bash
+set -e
+
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+OUTPUT_PATH=./dist/new-llama-awq/
+QUANT=q4f16_autoawq
+
+# python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT -o $OUTPUT_PATH --conv-template llama-2
+
+# python -m mlc_chat compile $OUTPUT_PATH -o $OUTPUT_PATH/llama.so
+
+python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH --source-format awq --source ../Llama-2-7B-Chat-AWQ/model.safetensors
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH --model-lib $OUTPUT_PATH/llama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/wasted/test_decode_awq_gemm.py b/wasted/test_decode_awq_gemm.py
new file mode 100644
index 0000000000..0fdd872694
--- /dev/null
+++ b/wasted/test_decode_awq_gemm.py
@@ -0,0 +1,110 @@
+# pylint: disable=invalid-name,missing-docstring
+from typing import List
+
+import numpy as np
+import pytest
+import safetensors
+import torch
+import tvm
+import tvm.testing
+from tvm import DataType
+from tvm.relax.frontend import nn
+
+from mlc_chat.compiler import QUANTIZATION
+from mlc_chat.compiler.loader import QuantizeMapping
+from mlc_chat.compiler.quantization import AWQQuantize
+from mlc_chat.support.auto_device import detect_device
+from mlc_chat.support.auto_target import detect_target_and_host
+
+
+def dequantize_np(
+    config: AWQQuantize,
+    weight: np.ndarray,
+    zeros: np.ndarray,
+    scale: np.ndarray,
+) -> np.ndarray:
+    def decode_int_arr(int_arr: np.ndarray, num_elem_per_storage: int, bits: int):
+        bin_mask = (1 << bits) - 1
+        int_arr_repeated = np.repeat(int_arr, num_elem_per_storage, axis=-1)
+        indice_j = np.indices(int_arr_repeated.shape)[1]
+        inv_order_map = [0, 4, 1, 5, 2, 6, 3, 7]
+        arr_bin = np.bitwise_and(
+            np.right_shift(
+                int_arr_repeated,
+                inv_order_map[indice_j % num_elem_per_storage] * bits,
+            ),
+            bin_mask,
+        )
+        return arr_bin
+
+    weight_bin = decode_int_arr(
+        weight, config.num_elem_per_storage, DataType(config.quantize_dtype).bits
+    )
+    zero_bin = decode_int_arr(
+        zeros, config.num_elem_per_storage, DataType(config.quantize_dtype).bits
+    )
+    scale_repeated = np.repeat(scale, config.group_size, axis=-1)
+    zero_bin_repeated = np.repeat(zero_bin, config.group_size, axis=-1)
+    return (weight_bin - zero_bin_repeated) * scale_repeated
+
+
+def test_dequantize_weight(quant_name: str, dtype: str):
+    target, _ = detect_target_and_host("auto")
+    device = detect_device("auto")
+    # print(target)
+    # print(device)
+    with safetensors.safe_open(
+        "../Llama-2-7B-AWQ/model.safetensors", framework="numpy", device="cpu"
+    ) as in_file:
+        q_proj_qweight = in_file.get_tensor("model.layers.0.self_attn.q_proj.qweight").astype(
+            "uint32"
+        )
+        q_proj_qzeros = in_file.get_tensor("model.layers.0.self_attn.q_proj.qzeros").astype(
+            "uint32"
+        )
+        q_proj_scales = in_file.get_tensor("model.layers.0.self_attn.q_proj.scales")
+    hf_state_dict = torch.load("/opt/models/llama-2/llama-2-7b-hf/pytorch_model-00001-of-00002.bin")
+    q_proj_weight = hf_state_dict["model.layers.0.self_attn.q_proj.weight"]
+    print(q_proj_weight)
+
+    class Test(nn.Module):
+        def __init__(self) -> None:
+            super().__init__()
+            self.linear = nn.Linear(
+                q_proj_weight.shape[1], q_proj_weight.shape[0], bias=False, dtype=dtype
+            )
+
+        def forward(self, x: nn.Tensor):
+            return self.linear(x)
+
+    config = QUANTIZATION[quant_name]
+    assert isinstance(config, AWQQuantize)
+    mod = config.quantize_model(Test(), QuantizeMapping({}, {}), "")
+    mod.linear.qweight.data = q_proj_qweight
+    mod.linear.qzeros.data = q_proj_qzeros
+    mod.linear.scales.data = q_proj_scales
+    with target:
+        model = mod.jit(
+            spec={
+                "forward": {
+                    "x": nn.spec.Tensor((q_proj_weight.shape[1], q_proj_weight.shape[1]), dtype)
+                }
+            },
+            pipeline="mlc_llm",
+            device="cuda:0",
+            target=target,
+        )
+    out = model["forward"](
+        torch.from_numpy(np.diag(np.ones(q_proj_weight.shape[1]).astype(dtype))).to(
+            "cuda:0"
+        )  # pylint: disable=no-member
+    )
+    print(out)
+    print(q_proj_weight.shape)
+    print(out.shape)
+    # out = dequantize_np(config, q_proj_qweight, q_proj_qzeros, q_proj_scales)
+    print(torch.max(torch.abs(out.detach().to("cpu") - q_proj_weight)))
+
+
+if __name__ == "__main__":
+    test_dequantize_weight("q4f16_awq_gemm", "float16")
diff --git a/wasted/test_mistral.py b/wasted/test_mistral.py
new file mode 100644
index 0000000000..895de58d1c
--- /dev/null
+++ b/wasted/test_mistral.py
@@ -0,0 +1,54 @@
+# pylint: skip-file
+import os
+import shutil
+
+import tvm
+
+tvm.error.InternalError
+
+temp_dir = "/opt/scratch/lesheng/mlc-llm/dist/tmp"
+
+os.environ["MLC_TEMP_DIR"] = temp_dir
+os.environ["MLC_CACHE_DIR"] = temp_dir
+
+import logging
+
+from mlc_chat import ChatConfig, ChatModule
+from mlc_chat.callback import StreamToStdout
+
+logging.basicConfig(
+    level=logging.INFO,
+    style="{",
+    datefmt="%Y-%m-%d %H:%M:%S",
+    format="[{asctime}] {levelname} {filename}:{lineno}: {message}",
+)
+
+import json
+
+
+def test_model(url):
+    os.makedirs(temp_dir, exist_ok=True)
+    cm = ChatModule(
+        url,
+        chat_config=ChatConfig(max_gen_len=10000, prefill_chunk_size=4096),
+        device="cuda",
+    )
+    output = cm.generate(
+        prompt="What is the meaning of life?",
+    )
+    print(output)
+    print(cm.stats())
+    for i in range(10):
+        prompt = input("User: ")
+        output = cm.generate(
+            prompt=prompt,
+        )
+        print(output)
+        print(cm.stats())
+    return output
+
+
+url = "HF://junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC"
+# url = "HF://junrushao/phi-2-q4f16_1-MLC"
+# url = "HF://junrushao/Llama-2-7b-chat-hf-q4f16_1-MLC"
+test_model(url)
diff --git a/wasted/test_mistral_multi_gpu.sh b/wasted/test_mistral_multi_gpu.sh
new file mode 100644
index 0000000000..440d0f6caf
--- /dev/null
+++ b/wasted/test_mistral_multi_gpu.sh
@@ -0,0 +1,19 @@
+QUANT=q4f16_1
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf
+OUTPUT_PATH=./dist/new-llama-${QUANT}/
+
+mkdir -p $OUTPUT_PATH
+
+python -m mlc_chat gen_config $MODEL_PATH/config.json --quantization $QUANT --conv-template llama-2 --num-shards $NUM_SHARDS -o $OUTPUT_PATH/params
+
+# python -m mlc_chat compile $OUTPUT_PATH/params -o $OUTPUT_PATH/llama.so
+
+# python -m mlc_chat convert_weight $MODEL_PATH --quantization $QUANT -o $OUTPUT_PATH/params
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH/params --model-lib $OUTPUT_PATH/llama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256 --num-shards 1
+
+# python -m mlc_chat compile --model $OUTPUT_PATH -o $OUTPUT_PATH/llama.so
+
+# python -m mlc_chat convert_weight --model $MODEL_PATH --quantization q4f16_1 -o $OUTPUT_PATH
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python -m mlc_chat.cli.benchmark --model $OUTPUT_PATH --model-lib $OUTPUT_PATH/llama.so --device "cuda:0" --prompt "What is the meaning of life?" --generate-length 256
diff --git a/wasted/test_mlc.py b/wasted/test_mlc.py
new file mode 100644
index 0000000000..38a5f812dd
--- /dev/null
+++ b/wasted/test_mlc.py
@@ -0,0 +1,30 @@
+import tvm
+
+from mlc_chat import ChatModule
+from mlc_chat.chat_module import ChatConfig
+
+
+@tvm.register_func("effect.print")
+def _print(_, array) -> None:
+    print(f"effect.print: shape = {array.shape}, dtype = {array.dtype}, data =\n{array}")
+
+
+import numpy as np
+
+array = tvm.nd.array(np.random.random((4, 20)))
+
+print(f"effect.print: shape = {array.shape}, dtype = {array.dtype}, data =\n{array}")
+
+# cm = ChatModule(
+#     # model="dist/llama-2-7b-chat-hf-q4f16_1/params",
+#     model="dist/new-llama-q4f16_1/params",
+#     chat_config=ChatConfig(
+#         temperature=0.01,
+#         top_p=0.9,
+#         max_gen_len=4096,
+#     ),
+#     model_lib_path="dist/new-llama-q4f16_1/llama.so",
+# )
+
+# output = cm.generate("What is the meaning of life?")
+# print(output)
diff --git a/wasted/test_mlc_hf.py b/wasted/test_mlc_hf.py
new file mode 100644
index 0000000000..e608048934
--- /dev/null
+++ b/wasted/test_mlc_hf.py
@@ -0,0 +1,23 @@
+import os
+
+os.environ["MLC_TEMP_DIR"] = "/opt/scratch/lesheng/mlc-llm/dist"
+os.environ["MLC_CACHE_DIR"] = "/opt/scratch/lesheng/mlc-llm/dist"
+
+import logging
+
+from mlc_chat import ChatModule
+from mlc_chat.callback import StreamToStdout
+
+logging.basicConfig(
+    level=logging.INFO,
+    style="{",
+    datefmt="%Y-%m-%d %H:%M:%S",
+    format="[{asctime}] {levelname} {filename}:{lineno}: {message}",
+)
+
+cm = ChatModule("HF://junrushao/Llama-2-7b-chat-hf-q3f16_1-MLC", device="cuda")
+output = cm.generate(
+    prompt="What is the meaning of life?",
+    # progress_callback=StreamToStdout(callback_interval=2),
+)
+print(output)
diff --git a/wasted/test_mlc_hf_batch.py b/wasted/test_mlc_hf_batch.py
new file mode 100644
index 0000000000..081f328153
--- /dev/null
+++ b/wasted/test_mlc_hf_batch.py
@@ -0,0 +1,81 @@
+# pylint: skip-file
+import os
+import shutil
+
+import tvm
+
+tvm.error.InternalError
+
+temp_dir = "/opt/scratch/lesheng/mlc-llm/dist/tmp"
+
+os.environ["MLC_TEMP_DIR"] = temp_dir
+os.environ["MLC_CACHE_DIR"] = temp_dir
+
+import logging
+
+from mlc_chat import ChatConfig, ChatModule
+from mlc_chat.callback import StreamToStdout
+
+logging.basicConfig(
+    level=logging.INFO,
+    style="{",
+    datefmt="%Y-%m-%d %H:%M:%S",
+    format="[{asctime}] {levelname} {filename}:{lineno}: {message}",
+)
+
+import json
+
+
+def test_model(url):
+    os.makedirs(temp_dir, exist_ok=True)
+    try:
+        cm = ChatModule(
+            url,
+            chat_config=ChatConfig(max_gen_len=10000, prefill_chunk_size=4096),
+            device="cuda",
+        )
+        output = cm.generate(
+            prompt="What is the meaning of life?",
+            # prompt="Write a python program that computes fibonacci sequence."
+        )
+        print(output)
+        for i in range(10):
+            output = cm.generate(
+                prompt="Talk more",
+                # prompt="Write a python program that computes fibonacci sequence."
+            )
+            print(output)
+    except:
+        output = "Failed"
+    # shutil.rmtree(temp_dir, ignore_errors=True)
+    return output
+
+
+url = "HF://junrushao/Mistral-7B-Instruct-v0.2-q4f16_1-MLC"
+output = test_model(url)
+print(output)
+exit(0)
+
+with open("model_info.json", "r") as f:
+    model_dict = json.load(f)
+
+testout_dir = "testout"
+os.makedirs(testout_dir, exist_ok=True)
+
+shutil.rmtree(temp_dir, ignore_errors=True)
+for model in model_dict:
+    model_infos = model_dict[model]
+    for quant in ["q3f16_1", "q4f16_1", "q4f32_1"]:
+        url = model_infos[quant]
+        if len(url) == 0:
+            continue
+        print(f"{model}\t{quant}: {url}")
+        testout_file = os.path.join(testout_dir, f"{model}_{quant}.out")
+        if f"{model}_{quant}" in ["CodeLlama-7b-hf_q4f32_1"]:
+            continue
+        if os.path.exists(testout_file):
+            print(f"skip {testout_file}")
+            continue
+        output = test_model(url)
+        with open(testout_file, "w+") as f:
+            f.write(output + "\n")
diff --git a/wasted/test_mlc_phi.py b/wasted/test_mlc_phi.py
new file mode 100644
index 0000000000..ed8a54b60f
--- /dev/null
+++ b/wasted/test_mlc_phi.py
@@ -0,0 +1,16 @@
+from mlc_chat import ChatModule
+from mlc_chat.chat_module import ChatConfig
+
+cm = ChatModule(
+    model="./dist/phi-2-q0f16/params",
+    model_lib_path="./dist/phi-2-q0f16/model.so",
+    # chat_config=ChatConfig(
+    #     temperature=0.01,
+    #     top_p=0.9,
+    #     max_gen_len=4096,
+    # ),
+)
+
+output = cm.generate("What is the meaning of life?")
+print(output)
+print(output)
diff --git a/wasted/test_mlc_red_pajama.py b/wasted/test_mlc_red_pajama.py
new file mode 100644
index 0000000000..43b708e7bb
--- /dev/null
+++ b/wasted/test_mlc_red_pajama.py
@@ -0,0 +1,14 @@
+from mlc_chat import ChatModule
+from mlc_chat.chat_module import ChatConfig
+
+cm = ChatModule(
+    model="dist/RedPajama-INCITE-Chat-3B-v1-q0f16/params",
+    chat_config=ChatConfig(
+        temperature=0.01,
+        top_p=0.9,
+        max_gen_len=4096,
+    ),
+)
+
+output = cm.generate("What is the meaning of life?")
+print(output)
diff --git a/wasted/test_old_red_pajama.sh b/wasted/test_old_red_pajama.sh
new file mode 100755
index 0000000000..63ce6a50db
--- /dev/null
+++ b/wasted/test_old_red_pajama.sh
@@ -0,0 +1,3 @@
+MODEL_PATH=/opt/scratch/lesheng/models/RedPajama-INCITE-Chat-3B-v1
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python build.py --model $MODEL_PATH --quantization q4f16_1 --use-cache 0 --debug-dump # --build-model-only
diff --git a/wasted/test_phi_hf.py b/wasted/test_phi_hf.py
new file mode 100644
index 0000000000..c738776e43
--- /dev/null
+++ b/wasted/test_phi_hf.py
@@ -0,0 +1,19 @@
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+torch.set_default_device("cuda")
+
+model = AutoModelForCausalLM.from_pretrained(
+    "/opt/scratch/lesheng/phi-2", torch_dtype="auto", trust_remote_code=True
+)
+tokenizer = AutoTokenizer.from_pretrained("/opt/scratch/lesheng/phi-2", trust_remote_code=True)
+
+inputs = tokenizer(
+    "What is the meaning of life?",
+    return_tensors="pt",
+    return_attention_mask=False,
+)
+
+outputs = model.generate(**inputs, max_length=200)
+text = tokenizer.batch_decode(outputs)[0]
+print(text)
diff --git a/wasted/test_phi_mlc.py b/wasted/test_phi_mlc.py
new file mode 100644
index 0000000000..482764cd26
--- /dev/null
+++ b/wasted/test_phi_mlc.py
@@ -0,0 +1,67 @@
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from tvm.relax.frontend import nn
+
+from mlc_chat.compiler.model.phi import modeling_phi as hf_phi
+from mlc_chat.compiler.model.phi import phi_model
+
+config = phi_model.PhiConfig()
+
+
+def test_rotary():
+    rotary_embed = phi_model.RotaryEmbedding(config)
+    print(config.n_head)
+    print(config.head_dim)
+    print(rotary_embed.position_embedding_base)
+    print(rotary_embed.rotary_dim)
+    mod = rotary_embed.jit(
+        spec={
+            "forward": {
+                "q": nn.spec.Tensor([1, 7, config.n_head, config.head_dim], "float16"),
+                "k": nn.spec.Tensor([1, 7, config.n_head, config.head_dim], "float16"),
+                "offset": int,
+            }
+        }
+    )
+    q = torch.rand(1, 7, config.n_head, config.head_dim, dtype=torch.float16)
+    k = torch.rand(1, 7, config.n_head, config.head_dim, dtype=torch.float16)
+    q_embed, k_embed = mod["forward"](q, k, 0)
+
+    hf_rotary = hf_phi.RotaryEmbedding(config.rotary_dim)
+    kv = torch.stack([k, k], dim=2)
+    hf_q, hf_kv = hf_rotary.forward(q, kv, 0)
+    print(torch.allclose(hf_q, q_embed))
+    # print(hf_q)
+    # print(q_embed)
+    hf_k, _ = torch.unbind(hf_kv, 2)
+    print(torch.allclose(hf_k, k_embed))
+    # print(q_embed.shape)
+
+
+def test_e2e():
+    hf_model = AutoModelForCausalLM.from_pretrained(
+        "/opt/scratch/lesheng/phi-2", torch_dtype=torch.float16, trust_remote_code=True
+    )
+    tokenizer = AutoTokenizer.from_pretrained("/opt/scratch/lesheng/phi-2", trust_remote_code=True)
+
+    state_dict = hf_model.state_dict()
+    state_dict["transformer.embd.weight"] = state_dict["transformer.embd.wte.weight"]
+    state_dict.pop("transformer.embd.wte.weight")
+    phi = phi_model.PhiForCausalLM(config)
+    phi.to("float16")
+    phi.load_state_dict(state_dict)
+    mod = phi.jit(spec=phi.get_default_spec())
+
+    inputs = tokenizer(
+        "What is the meaning of life?",
+        return_tensors="pt",
+        return_attention_mask=False,
+    )
+
+    mod["prefill"](inputs, len(inputs[0]))
+
+
+if __name__ == "__main__":
+    test_rotary()
+    # test_e2e()
+    # test_e2e()
diff --git a/wasted/test_preshard.sh b/wasted/test_preshard.sh
new file mode 100755
index 0000000000..504a80c719
--- /dev/null
+++ b/wasted/test_preshard.sh
@@ -0,0 +1,7 @@
+set -e
+
+MODEL_PATH=/opt/models/llama-2/llama-2-7b-chat-hf/
+
+CUDA_VISIBLE_DEVICES=4,5,6,7 python build.py --model $MODEL_PATH --quantization q4f16_1 --num-shards 2 --build-model-only --max-seq-len 4096 --use-cache 0 --use-presharded-weights
+
+# CUDA_VISIBLE_DEVICES=4,5,6,7 python build.py --model $MODEL_PATH --quantization q4f16_1 --num-shards 2 --convert-weight-only --max-seq-len 4096 --use-cache 0 --use-presharded-weights
diff --git a/wasted/worker.py b/wasted/worker.py
new file mode 100644
index 0000000000..59e2179016
--- /dev/null
+++ b/wasted/worker.py
@@ -0,0 +1,64 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+# pylint: disable=invalid-name
+"""Internal DiscoWorker for Disco ProcessSession."""
+import os
+import sys
+
+import numpy as np
+from tvm import register_func
+from tvm import runtime as _  # pylint: disable=unused-import
+from tvm._ffi import get_global_func
+
+from .. import base  # pylint: disable=unused-import, no-name-in-module
+
+
+def main():
+    """Main worker function"""
+    if len(sys.argv) != 5:
+        print("Usage: <worker_id> <num_workers> <read_fd> <write_fd>")
+        return
+
+    worker_id = int(sys.argv[1])
+    num_workers = int(sys.argv[2])
+    if sys.platform == "win32":
+        import msvcrt  # pylint: disable=import-outside-toplevel,import-error
+
+        reader = msvcrt.open_osfhandle(int(sys.argv[3]), os.O_BINARY)
+        writer = msvcrt.open_osfhandle(int(sys.argv[4]), os.O_BINARY)
+    else:
+        reader = int(sys.argv[3])
+        writer = int(sys.argv[4])
+
+    @register_func("vm.builtin.debug_print_disco")
+    def _print(lineo: str, array) -> None:
+        np_array = array.numpy()
+        np.save(f"disco_test/{worker_id}.npy", np_array)
+        print(f"{worker_id} done")
+        # print(
+        #     f"{worker_id}, {lineo}: shape = {array.shape}, dtype = {array.dtype}, data =\n{array}"
+        # )
+
+    worker_func = get_global_func("runtime.disco.WorkerProcess")
+    worker_func(worker_id, num_workers, reader, writer)
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except (KeyboardInterrupt, IOError):
+        pass