Repository navigation
ci(trtllm): bump CI wheel to TensorRT-LLM 1.3.0rc18 - #1663
Conversation
cp312 manylinux_2_28 x86_64 wheels for 1.3.0rc18 are published on pypi.nvidia.com (public PyPI stops at rc15; the script already installs via the nvidia extra index with --pre). The NCCL constraint stays as-is: upstream requirements.txt at v1.3.0rc18 pins the same nvidia-nccl-cu13>=2.28.9,<=2.29.2. CI-only: the Docker release pipelines (release-trtllm-docker.yml, nightly-engine-docker.yml) keep their rc10 base images. Signed-off-by: Simo Lin <25425177+slin1237@users.noreply.github.com>
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (1)
📝 WalkthroughWalkthroughThe CI installation script is updated to pin TensorRT-LLM pre-release wheel version ChangesTensorRT-LLM Version Upgrade
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~3 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
|
👋 The PR description doesn't fully follow PULL_REQUEST_TEMPLATE.md:
Please update the PR description so reviewers have the context they need. |
Workspace crates, bindings, helm chart, and release workflows bumped per make check-versions (smg 1.4.1 -> 1.5.0 plus per-crate minor/patch bumps). Also refresh the engine base images in the engine release and nightly docker workflows to the versions CI now tests against: - sglang: lmsysorg/sglang v0.5.10 -> v0.5.12.post1 (#1662) - vllm: vllm-openai matrix v0.19.0/v0.18.0/v0.17.0 -> v0.22.1/v0.21.0/v0.20.0 (#1625) - trtllm: NGC release matrix 1.3.0rc10/rc9/rc8 -> rc18/rc17/rc16 (#1663) All base image tags verified published (Docker Hub / nvcr.io). Signed-off-by: Simo Lin <25425177+slin1237@users.noreply.github.com>
Workspace crates, bindings, helm chart, and release workflows bumped per make check-versions (smg 1.4.1 -> 1.5.0 plus per-crate minor/patch bumps). Also refresh the engine base images in the engine release and nightly docker workflows to the versions CI now tests against: - sglang: lmsysorg/sglang v0.5.10 -> v0.5.12.post1 (#1662) - vllm: vllm-openai matrix v0.19.0/v0.18.0/v0.17.0 -> v0.22.1/v0.21.0/v0.20.0 (#1625) - trtllm: NGC release matrix 1.3.0rc10/rc9/rc8 -> rc18/rc17/rc16 (#1663) All base image tags verified published (Docker Hub / nvcr.io). Signed-off-by: Simo Lin <25425177+slin1237@users.noreply.github.com>
Description
CI-only bump of the TensorRT-LLM e2e wheel from 1.3.0rc14 to 1.3.0rc18. The Docker release pipelines (
release-trtllm-docker.yml,nightly-engine-docker.yml) are untouched and keep their rc10 base images.Changes (
scripts/ci_install_trtllm.shonly)TRTLLM_VERSION="1.3.0rc14"→"1.3.0rc18"NCCL_VERSION_CONSTRAINTdeliberately unchanged: upstreamrequirements.txtat tag v1.3.0rc18 pins the identicalnvidia-nccl-cu13>=2.28.9,<=2.29.2(and the sametorch>=2.10.0,<=2.11.0a0,cuda-python>=13).Test Plan
bash -n scripts/ci_install_trtllm.sh— clean.tensorrt_llm-1.3.0rc18-cp312-cp312-manylinux_2_28_x86_64.whlis published on pypi.nvidia.com (public PyPI stops at rc15; the script already resolves via--pre --extra-index-url https://pypi.nvidia.com, so the pin is installable as-is).Summary by CodeRabbit