[xinnywinne]

Summary

Adds speculative-decoding acceptance statistics to per-request metrics in the Rust OpenAI frontend.

This is a follow-up to #59314, which added the base per-request timing metrics support. It is not duplicate work: the Rust Chat Completions and Completions routes do not currently propagate the engine's per-request speculative-decoding statistics. This PR reuses the converter already used by /inference/v1/generate.

• exposes metrics.speculative_decoding for /v1/chat/completions and /v1/completions
• supports both non-streaming and streaming responses
• emits streaming metrics only on the final usage chunk
• keeps timing metrics and speculative-decoding metrics independently configurable
• reuses the existing speculative-decoding conversion shared with /inference/v1/generate

The response follows the existing Python/OpenAI per-request metrics schema, including summary and detailed speculative-decoding modes.

Validation

• cargo fmt --all -- --check
• cargo check --workspace --tests
• cargo clippy -p vllm-text -p vllm-chat -p vllm-server --all-targets -- -D warnings
• cargo test -p vllm-server routes::openai:: — 153 passed
• cargo test -p vllm-text --lib — 111 passed, 1 ignored
• cargo test -p vllm-chat --lib — 389 passed
• focused shared converter and propagation tests
• pre-commit hooks
• A100 end-to-end validation with n-gram speculative decoding:
  • Chat Completions, streaming and non-streaming
  • Completions, streaming and non-streaming
  • timing-only, speculative-decoding-only, and combined response behavior
  • intermediate streaming chunks omit metrics; the final usage chunk includes them

Notes

A broader cargo test -p vllm-text -p vllm-chat run also exercised the integration suites. Two model round-trip snapshot cases (roundtrip_mimo_v26 and roundtrip_step5) failed because the locally resolved model/tokenizer data no longer matched the checked-in snapshots; this PR only adds spec_decode_metrics: None to their terminal-output test fixture.

AI assistance was used to implement and validate this change. I reviewed every changed line and understand the implementation.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @xinnywinne.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[BugenZhao]

Thanks @xinnywinne

[BugenZhao]

/ci run

[github-actions]

✅ @xinnywinne, CI is now available for this PR.

• /ci run starts upstream CI; /amd-ci run starts AMD CI only.
• Run commands test your branch with every commit currently on its upstream target branch. If it is behind, the bot merges the target branch into it first (pull before pushing again). Append --allow-stale to a run command to test an outdated branch as is, at your own risk.
• /ci retry retries failed jobs in the CI build for the current PR head. If the current head has no CI build, it starts a new CI build for the current head containing only jobs that failed in the latest earlier CI build for this PR.
• /amd-ci retry retries failed jobs in AMD CI for the current PR head. Use /amd-ci run when the current head has no AMD CI build.
• /ci cancel cancels scheduled or running CI builds for this PR branch; /amd-ci cancel does the same for AMD CI only.

[github-actions]

✅ Triggered Buildkite CI #97623 for commit f42c7c387eb5.

ℹ️ This PR is 2 commits behind upstream main. Merged upstream main into this branch as f42c7c387eb5 before starting CI. Pull the branch before pushing more changes.