Repository navigation
Self-hosted models: remote code and tokenizer from the pinned revision; release 0.7.3 - #35
Merged
Merged
Conversation
…hts' pinned revision; release 0.7.3
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
A model that needs its own code (
trust_remote_code, Nemotron) loaded its code and tokenizer files from the Hub's default branch, not from the pinned revision of its weights. vLLM's log said so: transformers warned "A new version of the following files was downloaded ... To avoid downloading new versions of the code file, you can pin a revision" (seen on the Nemotron deploy of Nov 12).Now
vllm_command()adds--code-revisionand--tokenizer-revisionwith the profile's own revision whenevertrust_remote_codeis on. Both flags exist in vLLM 0.32.0 (checked with the CPU flag helper:--code-revision,--tokenizer-revision,--revisionall present). A profile without remote code (Qwen) is unchanged. Version 0.7.3.Tests (Windows, Node 22.13): Python settings tests 12/12 (new: the three revision flags carry the pinned revision; Qwen has none);
test/selfhost-modal.test.mjsandtest/model-provider.test.mjs19/19. Linux: this PR's GitHub job.