What's the difference between `LMMS_EVAL_USE_CACHE` and the `--use_cache` argument?

[pspdada]

Hi, I noticed there seem to be two different ways to enable caching in version 0.5, and I'm a bit confused about their differences.

The first method uses environment variables:

export LMMS_EVAL_USE_CACHE=True
export LMMS_EVAL_HOME="/path/to/cache_root"  # optional

python -m lmms_eval \
  --model async_openai \
  --model_args model_version=gpt-4o-2024-11-20,base_url=$OPENAI_API_BASE \
  --tasks mmmu_val \
  --batch_size 1 \
  --output_path ./logs/

The second method uses a command-line argument:

--use_cache "$CACHE_PATH"

According to the docstring:

:param use_cache: str, optional
A path to a sqlite db file for caching model responses. None if not caching.

Could you please clarify the distinction between these two caching mechanisms? Specifically:

• What does LMMS_EVAL_USE_CACHE control, and how does it interact with --use_cache?
• Is LMMS_EVAL_HOME related to the path provided in --use_cache, or are they independent?
• Which one should be preferred for caching model responses?

Thanks for your help!