Remove tensorrt-llm from optional-dependencies for uv compat#378
Remove tensorrt-llm from optional-dependencies for uv compat#378
Conversation
tensorrt-llm cannot be resolved by uv: the PyPI stub requires NVIDIA's extra index to build, and the NVIDIA wheel carries git-URL transitive deps (flashinfer, s2wrapper) that uv categorically rejects. Drop the extra and document manual pip install instead so that `uv sync --group dev` (and other non-TensorRT workflows) succeed. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f056b45f81
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| # tensorrt-llm is not declarable as a uv-resolvable extra because the PyPI | ||
| # stub needs NVIDIA's index and the NVIDIA wheel carries git-URL deps that | ||
| # uv rejects. Install manually: | ||
| # pip install --extra-index-url https://pypi.nvidia.com/ 'rank-llm[local]' 'tensorrt-llm[all]>=0.14.0' |
There was a problem hiding this comment.
tensorrt-llm extra defined for setup compatibility
Removing the tensorrt-llm optional dependency key makes existing setup commands fail immediately (for example, uv sync --group dev --extra tensorrt-llm now errors with “Extra tensorrt-llm is not defined”), which breaks documented TensorRT-LLM onboarding and any local/CI automation that still references that extra name. Even if TensorRT itself must be installed manually, preserving a compatibility extra (or updating all usage sites in the same change) avoids this regression.
Useful? React with 👍 / 👎.
1 participant
Summary
tensorrt-llmfrom[project.optional-dependencies]andallextratensorrt-llmis uv-incompatible: the PyPI stub needs NVIDIA's extra index, and the NVIDIA wheel carries git-URL transitive deps (flashinfer,s2wrapper) that uv rejectspip install --extra-index-url https://pypi.nvidia.com/ 'rank-llm[local]' 'tensorrt-llm[all]>=0.14.0'uv sync --group devnow succeeds without attempting to build tensorrt-llmTest plan
uv sync --group dev— passes, installs only dev toolinguv sync --group dev --extra local— passesuv sync --group dev --extra vllm— passesuv sync --group dev --extra sglang— passespip install -e .[all]behavior unchanged (minus tensorrt-llm)🤖 Generated with Claude Code