Skip to content

chore(frontend): bump vllm rust crates to cc706b05 - #898

Merged
xiaguan merged 1 commit into
pegainfer-project:mainfrom
RicardoMin:fix/vllm-pin-cc706b05
Aug 17, 2026
Merged

xiaguan merged 1 commit into
pegainfer-project:mainfrom
RicardoMin:fix/vllm-pin-cc706b05

Conversation

@RicardoMin

Copy link
Copy Markdown
Contributor

Summary

This PR updates the Rust vLLM dependencies required by the frontend and keeps the change separate from #865.

  • Bump the five directly pinned vLLM Rust crates from 8e61b646 to cc706b05.
  • Refresh Cargo.lock with the dependency closure required by the new revision.
  • Add EngineCoreOutput::ec_transfer_params: None for compatibility with the updated API.

Dependency Update

Crate Previous revision New revision
vllm-chat 8e61b646 cc706b05
vllm-engine-core-client 8e61b646 cc706b05
vllm-server 8e61b646 cc706b05
vllm-text 8e61b646 cc706b05
vllm-tokenizer 8e61b646 cc706b05

Validation

Scope Command Result
vLLM tokenizer tests cargo test --release --locked --features test-utils -p vllm-tokenizer --lib 57 passed, 0 failed
vLLM text tests cargo test --release --locked -p vllm-text --lib 70 passed, 0 failed, 1 ignored
PegaInfer frontend tests cargo test --release --locked -p pegainfer-frontend --lib 63 passed, 0 failed
PegaInfer frontend compilation cargo check --release --locked -p pegainfer-frontend --all-targets Passed
Patch formatting git diff --check upstream/main...HEAD Passed

The tokenizer test suite includes the zero-generated-token flush regressions:

Regression test Result
flush_without_push_token_does_not_leak_prompt Passed
flush_without_push_token_does_not_leak_undecodable_prompt_tail Passed

The single ignored vllm-text test requires network access to Hugging Face and is marked as too slow for CI.

Signed-off-by: RicardoMin <17879681016@163.com>
@xiaguan
xiaguan merged commit bfb64d6 into pegainfer-project:main Aug 17, 2026
10 of 13 checks passed
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.

2 participants