Your current environment

Your output of `python collect_env.py` here

🐛 Describe the bug

Hi, thanks for the great work on vLLM!

While reviewing the RunAI object storage integration, I noticed a potential inconsistency in maybe_pull_model_tokenizer_for_runai.

In the tokenizer branch:

object_storage_tokenizer = ObjectStorageModel(url=tokenizer)
object_storage_tokenizer.pull_files(
    model,
    ignore_pattern=[...],
)

Here, model is passed to pull_files() instead of tokenizer.

From the RunAI Model Streamer S3 implementation:

def pull_files(model_path: str, dst: str, ...):
    ...
    bucket_name, base_dir, files = list_files(s3, model_path, ...)
    ...
    s3.download_file(bucket_name, file, destination_file)

The model_path argument is directly used as the object storage source for listing and downloading files.

This suggests that in the tokenizer branch above, files would be pulled from model rather than tokenizer.

This seems inconsistent with the expected behavior where model and tokenizer URIs are handled independently.

When model != tokenizer, this may lead to:

• tokenizer files being fetched from the model path
• missing or incorrect tokenizer files
• silent mismatch between model and tokenizer

This issue is likely masked when model == tokenizer.

The tokenizer branch would likely call:

object_storage_tokenizer.pull_files(tokenizer, ...)

so that files are fetched from the correct object storage path.

Could you confirm whether this is expected behavior or a leftover issue?

Happy to submit a PR if this is unintended.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.