Skip to content

build(vllm): [cherry-pick 1.7.0] bump core to v0.33.0 and Omni to v0.33.0rc1 (#17862) - #18271

Merged
nv-nmailhot merged 1 commit into
release/1.7.0from
jdarve/cherry-pick-vllm-0.33-bump
Nov 6, 2026
Merged

nv-nmailhot merged 1 commit into
release/1.7.0from
jdarve/cherry-pick-vllm-0.33-bump

Conversation

@JulienDarve

Copy link
Copy Markdown
Contributor

Refs DYN-5386
Refs DYN-5387
Refs DYN-5388

Cherry-pick of #17862 onto release/1.7.0. The commit is the merged main commit f16353af7a, picked with git cherry-pick -x --signoff.

Summary

Moves the CUDA vLLM image to vLLM v0.33.0 and vLLM-Omni v0.33.0rc1, with the compatibility changes and the vLLM sidecar structural-tag default from #17862. The CPU and XPU images keep their pins. This clears three bugs reported against 1.7.0-rc.0:

  • DYN-5386: vllm-rs 0.32 reads every image as black; vLLM 0.33.0 fixes it (vllm#61586).
  • DYN-5387: every LoRA load through vllm-rs 0.32 fails; vLLM 0.33.0 fixes it (vllm#63295).
  • DYN-5388: Qwen3-TTS crashes at startup when stage 0 uses FlashInfer; Omni 0.33.0rc1 fixes the warmup (vllm-omni#9264), and the installer's Physis workaround keeps requests working on that backend.

The known issues are those listed in #17862: the CUDA image removes Omni's experimental Physis pipelines until vllm-omni#9939 is fixed, Omni video models still need the PyAV opt-in, LMCache's native ops do not load under torch 2.14, and the image pins Transformers 5.19.0.

Validation

  • The pick applies without conflicts. git patch-id --stable gives the same ID for f16353af7a and for this commit: 398cc38a05.
  • Since the branch cut, build(deps): pin AISim 0.14.0 RC0 sources for release/1.7.0 #18014 is the only release commit that touches files this PR changes (Cargo.lock, lib/bindings/python/Cargo.lock and pyproject.toml). They merge automatically, and cargo metadata --locked passes at the repository root and in lib/bindings/python.
  • docs/fern/scripts/gen_llms_tables.py --check passes.
  • pre-commit run on the 20 changed files passes. pytest-marker-report was skipped because it fails in the local environment on unchanged files too.
  • On main, build(vllm): bump core to v0.33.0 and Omni to v0.33.0rc1 #17862 passed full CI at 6988c5c858, and its description lists the local GPU validation.
  • Not run on this branch: image builds, GPU tests and the Rust tests (the local protoc 3.12 cannot build dynamo-llm). This PR's CI covers them.

Generated with Claude Code

Signed-off-by: Julien Darve <jdarve@NVIDIA.com>
(cherry picked from commit f16353a)
Signed-off-by: Julien Darve <jdarve@NVIDIA.com>
@JulienDarve
JulienDarve requested review from a team as code owners November 5, 2026 23:07
@github-actions github-actions Bot added build documentation Improvements or additions to documentation backend::vllm Relates to the vllm backend frontend `python -m dynamo.frontend` and `dynamo-run in=http|text|grpc` container labels Nov 5, 2026
@github-actions

github-actions Bot commented Nov 5, 2026

Copy link
Copy Markdown
Contributor

@nv-nmailhot
nv-nmailhot merged commit a942317 into release/1.7.0 Nov 6, 2026
143 of 145 checks passed
@nv-nmailhot
nv-nmailhot deleted the jdarve/cherry-pick-vllm-0.33-bump branch November 6, 2026 00:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

backend::vllm Relates to the vllm backend build container documentation Improvements or additions to documentation frontend `python -m dynamo.frontend` and `dynamo-run in=http|text|grpc` size/L

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants