chore: upgrade vLLM to 0.18.0 and torch/unsloth to equivalent versions#342
chore: upgrade vLLM to 0.18.0 and torch/unsloth to equivalent versions#342mckornfield wants to merge 1 commit intomainfrom
Conversation
mckornfield
commented
Apr 2, 2026
- Upgrade various versions to address CVEs
- Remove deprecated parameters
There was a problem hiding this comment.
Pull request overview
This PR upgrades the vLLM/Torch/Unsloth dependency stack to newer versions to address CVEs and adjusts the vLLM backend implementation/tests to align with removed/deprecated vLLM parameters and behaviors.
Changes:
- Upgraded vLLM to
0.18.0and updated Torch/TorchVision/TorchAO/Unsloth/XFormers pins accordingly. - Removed per-request logits processor usage from the vLLM backend (including removing the TypicalLogitsWarper wrapper) and updated unit tests to reflect that
typical_pis excluded. - Increased an end-to-end test timeout to accommodate longer runtimes with the upgraded stack.
Reviewed changes
Copilot reviewed 4 out of 5 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| tests/generation/test_vllm_backend.py | Updates mapping/unit tests to assert typical_p is excluded instead of creating a logits processor. |
| tests/e2e/test_safe_synthesizer.py | Extends e2e timeout to reduce flakiness under slower runs. |
| src/nemo_safe_synthesizer/generation/vllm_backend.py | Removes deprecated/unsupported vLLM request-time logits processing and deprecated structured-output args. |
| pyproject.toml | Bumps dependency pins for torch/torchvision/torchao/unsloth/vllm/xformers. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
a87679c to
01acad9
Compare
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 4 out of 5 changed files in this pull request and generated 1 comment.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
/sync |
4efd951 to
45b6003
Compare
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 3 out of 4 changed files in this pull request and generated no new comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
45b6003 to
514c635
Compare
nina-xu
left a comment
There was a problem hiding this comment.
i think this could work as is. approving pending a full slurm sweep
Signed-off-by: Matt Kornfield <mkornfield@nvidia.com>
514c635 to
6573f17
Compare
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 3 out of 4 changed files in this pull request and generated no new comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.