fix(parser): prevent DSML markup leak in DeepSeek V4 streaming tool calls

[SparshM8]

Description

Fixes #53227.

Streaming tool calls on DeepSeek-V4-Pro leak raw DSML markup into tool call arguments when using the deepseek_v4 tool call parser. This happens because _PARTIAL_PARAM_RE uses a greedy (.*)$ match that captures trailing DSML tags (like </｜DSML｜parameter) when they are only partially received or processed during streaming.

Changes

• Updated _PARTIAL_PARAM_RE in vllm/parser/deepseek_v4.py to use a non-greedy match with a negative lookahead ((?:(?!</?{_ESCAPED_DSML}).)*). This ensures the parser stops capturing argument content as soon as it encounters the start of any DSML tag.

Testing

Verified with a standalone reproduction script that simulates partial streaming chunks:

raw = '<｜DSML｜parameter name="location" string="true">Paris</｜DSML｜parameter'
# Before: {"location": "Paris</｜DSML｜parameter"}
# After: {"location": "Paris"}

Confirmed that the fix correctly handles both complete and partial DSML tags without leaking markup into the argument JSON.

[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.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

PRs do not trigger a full CI run by default. Reviewers with write access and configured trusted contributors can comment /ci run for upstream CI or /amd-ci run for AMD CI only whenever CI signals are needed.

Once the PR is approved or has the ready label, the PR author can also use the corresponding /ci run, /ci retry, and /ci cancel commands, or their /amd-ci variants. New commits do not start upstream CI automatically.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

Agent Guidelines

IMPORTANT: 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.

🚀

[SparshM8]

Closing in favor of #53405 which provides a more comprehensive fix and better test coverage.