Loading base model/lora adapters from s3 fails and throws HF error

[robert-moyai]

🐛 Describe the bug

Im trying to initialize AIBrix using model files hosted on s3. But I am not able to get vllm-openai container to run as its incorrectly identifying the s3 link as being a HF repo.

huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': 's3://s3-bucket/llama-3.1-nemoguard-8b-topic-control'. Use repo_type argument if needed.

Steps to Reproduce

1. Move a model llama-3.1-nemoguard-8b-topic-control to a s3 bucket
2. Deploy AIBrix following the docs: https://aibrix.readthedocs.io/latest/features/lora-dynamic-loading.html#create-base-model. Be sure to reference the model using its s3 path: s3://s3-bucket/llama-3.1-nemoguard-8b-topic-control
3. Inspect the pod logs and find the error. Pod will crash when applying.
4. Same issue if you just load LoRA from s3

Expected behavior

Pod should just start running like when downloading model from HF

Environment

• AIBrix version: 0.20.0
• k8s deployment
• AWS s3 bucket with model files