Feature: Support downloading model weights on-the-fly from HuggingFace (#166)#167
Merged
XkunW merged 24 commits intoVectorInstitute:mainfrom Mar 30, 2026
Merged
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #167 +/- ##
==========================================
- Coverage 91.23% 89.97% -1.27%
==========================================
Files 14 14
Lines 1438 1466 +28
==========================================
+ Hits 1312 1319 +7
- Misses 126 147 +21
🚀 New features to boost your workflow:
|
XkunW
reviewed
Nov 18, 2025
Contributor
XkunW
left a comment
XkunW
left a comment
There was a problem hiding this comment.
Hi @rohan-uiuc, thanks for opening this, I left a few comments. Another thing worth considering is adding a check in the API to see if a model needs to be downloaded, and if that's the case, only allow the download if the HF cache directory env var is set, so that users wouldn't accidentally download a model to their home directory and use up all the quota
Adds hf_model field to ModelConfig and LaunchOptions to specify a HuggingFace model id for vLLM to download at runtime.
Updates SlurmScriptGenerator and BatchSlurmScriptGenerator to use hf_model for vllm serve when local weights don't exist. Priority: local weights > hf_model > model name.
XkunW
approved these changes
Mar 30, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR Type
Feature
Short Description
Implements support for on-the-fly model weight downloads from HuggingFace when local model weights directory doesn't exist. This allows users to launch models without manually downloading and mounting weight directories.
The code now checks if the model weights directory exists before attempting to bind mount it. If the directory doesn't exist, it skips the bind mount and uses the model identifier from
--modelin vllm_args (or falls back to model_name). Users must pass the full HuggingFace model identifier (e.g.,Qwen/Qwen2.5-7B-Instruct) via--modelin vllm_args for automatic downloads to work.Fixes #166
Tests Added
test_generate_server_setup_singularity_no_weights: Verifies server setup doesn't include model weights path when directory doesn't existtest_generate_launch_cmd_singularity_no_local_weights: Verifies launch command uses HF model identifier when local weights are missingtest_generate_model_launch_script_singularity_no_weights: Verifies batch mode correctly handles missing model weights--modelis specified in vllm_args