Integration for Ray LLM with load_format=runai_streamer

[jiangwu300]

Purpose

This PR makes it so that when a model is a s3 (or other cloud storage location) and we have runai_streamer turned on, we do not overwrite the self.model with a local directory which leads to failure of engine initialization since the local dir will not have the safe tensors.

Test Plan

Run a simple ray job with ray data llm, passing in a s3 model path with runai_streamer turned on.

Test Result

Runs as expected, no more failure.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors.

You ask your reviewers to trigger select CI tests on top of fastcheck CI.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

[gemini-code-assist]

Code Review

This pull request addresses a bug where the model path is incorrectly overwritten when using runai_streamer with models from cloud storage. The fix prevents this overwrite, ensuring the engine initializes correctly. My review identifies an oversight where the fix doesn't account for the runai_streamer_sharded load format, which would lead to the same bug. I've suggested a more robust condition to cover all runai_streamer variants.

[gemini-code-assist]

The condition to prevent overwriting self.model only checks for load_format=="runai_streamer". This is incomplete as it misses other runai_streamer variants like runai_streamer_sharded. If a user specifies load_format="runai_streamer_sharded" with a cloud storage model path, self.model will be incorrectly overwritten with a local directory, leading to the same failure this PR aims to fix. Using startswith("runai_streamer") would be more robust and cover all related formats.

Suggested change

	if not (is_cloud_storage(self.model) and self.load_format=="runai_streamer"):
	if not (is_cloud_storage(self.model) and self.load_format.startswith("runai_streamer")):

[jiangwu300]

Are there other modes we need to consider as well like tokenizer? I'm not sure which load_formats actually skip tensors being downloaded and use streaming

[mergify]

Hi @jiangwu300, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

[Isotr0py]

After a second eye, I think it's better to fix this inside ModelConfig where redirected model, WDYT?

vllm/vllm/config/model.py

Lines 769 to 787 in 02a4169

	def maybe_pull_model_tokenizer_for_runai(self, model: str, tokenizer: str) -> None:
	"""Pull model/tokenizer from Object Storage to temporary
	directory when needed.
	Args:
	model: Model name or path
	tokenizer: Tokenizer name or path
	"""
	if not (is_runai_obj_uri(model) or is_runai_obj_uri(tokenizer)):
	return
	if is_runai_obj_uri(model):
	object_storage_model = ObjectStorageModel(url=model)
	object_storage_model.pull_files(
	model, allow_pattern=["*.model", "*.py", "*.json"]
	)
	self.model_weights = model
	self.model = object_storage_model.dir