Allow vLLM 0.25–0.32 without exact runtime pins

#1
by emelryan - opened

Allow stable vLLM >=0.25.0,<0.33 in the dependency and runtime checks. Relax the Torch, torchvision, Transformers, Hub, and Pillow pins needed by those releases; declare packaging for range checks. Only pyproject.toml and vllm_tied_patch/plugin.py change.

Validation: source/API checks and Linux dependency dry-runs passed for all ten stable releases from 0.25.0 through 0.32.0; runtime range checks passed. GPU inference has not been tested.

emelryan changed pull request status to merged

Sign up or log in to comment