From f4f932aae6ab1baab3f5c856b6409db7cefbef85 Mon Sep 17 00:00:00 2001 From: Arjun Date: Fri, 23 Feb 2024 21:08:06 +0000 Subject: [PATCH] Updated dl2q edge configs --- cm-mlops/script/compile-model-for.qaic/_cm.json | 4 ++-- .../customize.py | 2 +- .../reproduce-mlperf-inference-qualcomm/_cm.yaml | 16 +++++++++++++--- 3 files changed, 16 insertions(+), 6 deletions(-) diff --git a/cm-mlops/script/compile-model-for.qaic/_cm.json b/cm-mlops/script/compile-model-for.qaic/_cm.json index 472f315b4c..ca305b860d 100644 --- a/cm-mlops/script/compile-model-for.qaic/_cm.json +++ b/cm-mlops/script/compile-model-for.qaic/_cm.json @@ -305,12 +305,12 @@ }, "bert-99.9,server": { "env": { - "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=4" + "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=2" } }, "bert-99.9,server,nsp.14": { "env": { - "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=4" + "CM_QAIC_MODEL_COMPILER_ARGS_SUT": "-aic-num-cores=2" } }, "tf": { diff --git a/cm-mlops/script/generate-mlperf-inference-submission/customize.py b/cm-mlops/script/generate-mlperf-inference-submission/customize.py index dccd41e54f..c29b34e241 100644 --- a/cm-mlops/script/generate-mlperf-inference-submission/customize.py +++ b/cm-mlops/script/generate-mlperf-inference-submission/customize.py @@ -59,7 +59,7 @@ def generate_submission(i): system_meta['division'] = env['CM_MLPERF_SUBMISSION_DIVISION'] if 'CM_MLPERF_SUBMISSION_CATEGORY' in env: - system_meta['system_type'] = env['CM_MLPERF_SUBMISSION_CATEGORY'] + system_meta['system_type'] = env['CM_MLPERF_SUBMISSION_CATEGORY'].replace("-", ",") duplicate= (env.get('CM_MLPERF_DUPLICATE_SCENARIO_RESULTS', 'no') in ["yes", "True"]) diff --git a/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml b/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml index afe9f1d24d..e4a13ac6dd 100644 --- a/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml +++ b/cm-mlops/script/reproduce-mlperf-inference-qualcomm/_cm.yaml @@ -484,7 +484,7 @@ variations: env: CM_MODEL: bert-99 CM_SQUAD_ACCURACY_DTYPE: float32 - CM_NOT_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3750364/files/bert_large_v1_1_fake_quant.onnx" + CM_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3750364/files/bert_large_v1_1_fake_quant.onnx" bert-99.9: group: model @@ -492,7 +492,7 @@ variations: - bert_ env: CM_MODEL: bert-99.9 - CM_NOT_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3733910/files/model.onnx" + CM_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://zenodo.org/record/3733910/files/model.onnx" loadgen-batch-size.#: group: loadgen-batch-size @@ -611,6 +611,10 @@ variations: qaic_activation_count: "3" dl2q.24xlarge,bert-99.9,offline: + env: + qaic_activation_count: "7" + + dl2q.24xlarge,bert-99,offline: env: qaic_activation_count: "14" @@ -626,7 +630,7 @@ variations: dl2q.24xlarge,bert-99.9,server: env: - qaic_activation_count: "14" + qaic_activation_count: "7" dl2q.24xlarge,retinanet,server: default_variations: @@ -634,6 +638,12 @@ variations: env: qaic_activation_count: "14" + dl2q.24xlarge,resnet50,multistream: + default_variations: + batch-size: bs.1 + env: + qaic_activation_count: "1" + pro: env: qaic_queue_length: 10