Skip to content

feat(vllm): Bump vllm-tensorizer to vLLM v0.18.1rc0#139

Closed
alexeldeib wants to merge 1 commit intomainfrom
ace/vllm-v0181rc0-pr
Closed

feat(vllm): Bump vllm-tensorizer to vLLM v0.18.1rc0#139
alexeldeib wants to merge 1 commit intomainfrom
ace/vllm-v0181rc0-pr

Conversation

@alexeldeib
Copy link
Copy Markdown
Contributor

@alexeldeib alexeldeib commented Mar 24, 2026

Summary

  • Bumps vLLM from v0.18.0 to v0.18.1rc0
  • Includes vllm#37158 which fixes FakeTensorMode AttributeError on Python 3.10 + torch 2.10.0
  • K2.5 NVFP4 + EAGLE3 speculative decoding (n=4)
  • Server starts cleanly through torch.compile/CUDA graph capture without FakeTensorMode errors

@alexeldeib alexeldeib requested a review from a team as a code owner March 24, 2026 21:29
Includes vllm#37158 which fixes FakeTensorMode AttributeError on
Python 3.10 + torch 2.10.0 (standalone_compile module resolution).
@alexeldeib alexeldeib force-pushed the ace/vllm-v0181rc0-pr branch from 9080ea6 to 75990bd Compare March 24, 2026 21:32
@alexeldeib alexeldeib marked this pull request as draft March 24, 2026 22:01
@alexeldeib
Copy link
Copy Markdown
Contributor Author

leaving as draft, v0.18 has sadness but can wait for v0.18.1 to cut a real release

@github-actions
Copy link
Copy Markdown

@alexeldeib Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/23513224200
Image: ghcr.io/coreweave/ml-containers/vllm-tensorizer:ace-vllm-v0181rc0-pr-75990bd-v0.18.1rc0

@alexeldeib alexeldeib closed this Mar 26, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant