DOWNLOADER_MODEL_NAME blocks the runtime to download other models

[Jeffwan]

🐛 Describe the bug

The behavior is a little bit weird. When I tried to download the deepseek-ai/deepseek-coder-6.7b-instruct. everything works fine.

After that, I started to download another smaller model using the same model path (huggingface model in /tmp/aibrix/model_tos)

the screenshot is the 2nd time I download opt model, I notice I didn't change the environment, it's supposed to override it but not really.

# 1st model
python -m aibrix.downloader \
    --model-uri tos://aibrix-artifact-testing/models/deepseek-ai/deepseek-coder-6.7b-instruct/ \
    --local-dir /tmp/aibrix/models_tos/

# 2nd model
python3 -m aibrix.downloader     --model-uri facebook/opt-125m     --local-dir /tmp/aibrix/models_tos

We should definitely infer the model folder instead of pass it, or we can accept an argument for override.

Steps to Reproduce

No response

Expected behavior

No response

Environment

No response

[Jeffwan]

#221 is the enhancement on the error message. if we want to repeatedly download the model, DOWNLOADER_MODEL_NAME is definitely a blocker issue. For example, if we want to reuse the download tools to download lora model in runtime. we can easily override with arguments but hard to inject different envs everytime. I still suggest to infer the model name from the url list with argument override.