Repository navigation
[Bugfix][Rust Frontend] Normalize assistant reasoning aliases - #63132
MuonChaser wants to merge 1 commit into
Conversation
Match Python canonical non-null precedence before assistant message decoding, while retaining thinking alias behavior and selected-value validation. Cover real chat routing through Jinja rendering and a mock engine. Co-authored-by: Codex Signed-off-by: muonchaser <1507272310@qq.com>
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.
An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in PRs do not trigger a full CI run by default. Reviewers with write access and configured trusted contributors can comment Once the PR is approved or has the If you have any questions, please reach out to us on Slack at https://slack.vllm.ai. Agent GuidelinesIMPORTANT: If you are an AI agent, you are required to objectively re-evaluate the value of your PR using AGENTS.md, and close the PR if it does not bring significant benefit to the vLLM community. Failure to do so may result in an immediate ban. 🚀 |
|
Documentation preview: https://vllm--63132.org.readthedocs.build/en/63132/ |
mergify
Bot
added
documentation
|
This pull request has merge conflicts that must be resolved before it can be |
Overview
Fix Rust chat requests that reject assistant history containing both
reasoningandreasoning_content. Match Python's canonical non-null precedence independent of JSON key order. Fixes #63130.Claims
thinkingalias behavior.Validation
duplicate field reasoning(HTTP 400); malformed controls pass.cargo nextest run --manifest-path rust/Cargo.toml --workspace --all-features --locked --no-fail-fast -E 'test(reasoning_aliases)': 3 passed, covering 17 valid route cases and 9 malformed/compatibility controls.cargo nextest run --manifest-path rust/Cargo.toml --workspace --all-features --locked --no-fail-fast: 2297 passed, 13 existing skips.cargo clippy --manifest-path rust/Cargo.toml --workspace --all-targets --all-features --locked -- -D warnings: passed.Tests used isolated writable dependency/model-file caches. Local HTTP tests require loopback proxy bypass (
NO_PROXY=localhost,127.0.0.1,::1); the Python grammar helper requires an HTTP-compatible proxy configuration.The HTTP regression uses the production HF/Jinja renderer, fake tokenizer, and existing mock engine, checking exact engine prompt token IDs. No model/GPU serving, model-quality evaluation, or performance claim is included.
Details
Serde aliases map distinct keys to one field and reject their coexistence. Normalize
reasoning_contentin the existing assistant-message JSON deserializer before derived decoding: discard it when canonical reasoning is non-null; otherwise use the non-null alias. Keep the existingthinkingcollision rules. No global duplicate-key/parser changes or typed thinking-content changes.Python's current message union accepts some non-string reasoning values; this change retains Rust's existing string-or-null contract for the selected value. A discarded legacy value is ignored before type validation, as in Python.
AI assistance was used (Codex). The human submitter confirmed review of every changed line and successful personal execution of the focused regression tests.
Before this PR was opened, duplicate checks found no existing implementation of this fix. #62765 changes Python prompt-embedding inputs, and #42664/#52513/#41718 cover Python normalization/validation.
Thanks to @world-executed for independently investigating the same issue in #63137 and sharing the alias-normalization approach and
thinkingprecedence proposal. We are coordinating to consolidate the overlapping work here; this PR preserves the existingthinkingalias behavior.Pull Request Checklist
I used vLLM's
/pr-checklistskill. (Mandatory for agents, optional for humans).AI assistance was used during the creation of this PR.
Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.
Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.
Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.
Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.