Issue about preparing checkpoint.

[Rexzhan]

Hi~really nice work! I encountered a problem following your inference pipeline.
Specifically, i downloaded llama checkpoint from https://huggingface.co/huggyllama/llama-7b, and used the command:
python3 -m llava.model.apply_delta --base /path/to/llama-7b --target /output/path/to/LLaVA-7B-v1 --delta liuhaotian/LLaVA-Lightning-7B-delta-v1-1 and modified the apply_delta.py by claiming safe_serialization=False
to prepare the 7b pytorch llava weights, and it turned out to be three bin files instead of two, which is inconsistent with \Line 190 in ./test/DS_SmartEdit_test.py:
LLaVA_00002_weights = LLaVA_model_path + "/pytorch_model-00002-of-00002.bin".

Could you please tell me what i am doing wrong~~

[EnableAsync]

Hello, I also encountered the same problem, and it was solved after modifying the max_shard_size parameter to 10GB.