Remove unnecessary numpy/thinc/spacy version overrides#243
Closed
smirnovlad wants to merge 1 commit into
Closed
Conversation
The numpy>=2.0 constraint, thinc>=8.3, and spacy>=3.8 pins were based on the assumption that vLLM requires numpy 2.x. Verified that vLLM 0.12.0 works fine with numpy 1.26.4, thinc 8.2.5, and spacy 3.7.5. Changes: - pyproject.toml: numpy>=2.0.0,<2.3.0 -> numpy>=1.23.5, remove thinc pin - setup.sh: remove post-install numpy/thinc/spacy pinning block, remove spacy sed patch, update transformers sed to match main branch
Collaborator
Author
|
Closing — all changes here are superseded by #245, which includes the numpy/thinc cleanup plus additional dependency improvements (lm-polygraph as direct dep, vLLM moved to main deps, bundled latex2sympy). |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Remove dependency overrides that were based on the incorrect assumption that vLLM requires numpy 2.x.
Verified that the current working environment uses numpy 1.26.4, thinc 8.2.5, spacy 3.7.5, and vLLM 0.12.0 — all tests pass (195/195), eval pipeline works end-to-end.
Changes
pyproject.toml:
numpy>=2.0.0,<2.3.0→numpy>=1.23.5(no upper bound needed)thinc>=8.3.0(not needed with numpy 1.x)setup.sh:
numpy>=2.0,<2.3+thinc>=8.3+spacy>=3.8pinning blockspacysed patch on lm-polygraphtransformerssed patch to match current lm-polygraph main (==4.50.0→>=4.50.0)Why this matters for PyPI
Every override in
setup.shis a hack thatpip install thinkboostercan't reproduce. Removing unnecessary ones brings us closer to a clean PyPI package.Test plan
pytest tests/ -v— 195 passed