[i-square]

Purpose

Add dedicated intern-s1 reasoning and tool parsers without changing the
existing internlm parser behavior.

This keeps the fix scoped to Intern-S1 and avoids compatibility risk for
existing internlm users and models.

This PR includes:

• a new intern-s1 reasoning parser that hoists action blocks out of
  <think>...</think> into content for non-streaming responses, so the
  tool parser can extract standard OpenAI tool calls without changing the
  shared serving flow
• a new intern-s1 tool parser that is more tolerant of the current
  Intern-S1 output shape:
  • allows whitespace/newlines between <|action_start|> and <|plugin|>
  • supports multiple action blocks in non-streaming extraction
  • falls back to plain text when tool JSON is malformed instead of raising
  • adds a safer fallback for the first streamed arguments delta
• lazy registration for both --reasoning-parser intern-s1 and
  --tool-call-parser intern-s1
• focused unit tests for the new parsers

The intent is to fix Intern-S1 tool parsing with the smallest possible impact
surface, rather than changing generic InternLM parsing behavior for all users.

Test Plan

Static checks run locally:

git diff --check
python -m compileall \
  vllm/tool_parsers/intern_s1_tool_parser.py \
  vllm/reasoning/intern_s1_reasoning_parser.py \
  tests/tool_parsers/test_intern_s1_tool_parser.py \
  tests/reasoning/test_intern_s1_reasoning_parser.py

Targeted tests added in this PR:

pytest tests/tool_parsers/test_intern_s1_tool_parser.py
pytest tests/reasoning/test_intern_s1_reasoning_parser.py

The new tests cover:

• intern-s1 parser registration
• hoisting action blocks from reasoning to content
• keeping deepseek_r1 behavior when no action block is present
• non-streaming parsing with spaced special tokens
• non-streaming parsing with multiple action blocks
• graceful fallback on malformed JSON
• fallback behavior for the initial streamed arguments delta

Test Result

Passed locally:

• git diff --check
• python -m compileall ... for the new parser and test files
• python -m pytest tests/tool_parsers/test_intern_s1_tool_parser.py tests/reasoning/test_intern_s1_reasoning_parser.py -q
  • result: 9 passed

Not run locally:

• ruff

Notes about the local test environment:

• pytest and tblib were installed into the local test env
• mistral-common was aligned to 1.11.0 to match the repository's test
  dependency set
• the new test files were made self-contained so they do not rely on external
  tokenizer downloads or GPU cleanup during teardown

Documentation update:

• Updated docs/features/tool_calling.md to recommend
  --tool-call-parser intern-s1 --reasoning-parser intern-s1 for Intern-S1
  models instead of reusing the generic internlm parser guidance.
• Updated docs/features/reasoning_outputs.md to list Intern-S1 as a
  supported reasoning model with tool calling and to show the recommended
  server flags.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

[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.

[mergify]

Documentation preview: https://vllm--40115.org.readthedocs.build/en/40115/

[mergify]

Hi @i-square, the pre-commit checks have failed. Please run:

uv pip install pre-commit>=4.5.1
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy failing?

mypy is run differently in CI. If the failure is related to this check, please use the following command to run it locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10

[gemini-code-assist]

Code Review

This pull request introduces support for Intern-S1 and Intern-S1-mini models, adding dedicated reasoning and tool-calling parsers. The implementation includes the InternS1ReasoningParser, which hoists action blocks from reasoning to content, and the InternS1ToolParser for extracting tool calls in both streaming and non-streaming modes. Documentation and tests for these new features are also provided. However, critical bugs were identified in the streaming tool parser implementation: one logic error prematurely blocks the streaming of tool arguments, and another issue involves incorrect list indexing that would lead to IndexError exceptions during execution.

[i-square]

@aarnphm @chaunceyjiang @sfeng33 gentle ping on this PR. Let me know if you need any additional context or changes to make the review easier. Thanks!

[github-actions]

This pull request has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this pull request should remain open. Thank you!

[github-actions]

This pull request has been automatically closed due to inactivity. Please feel free to reopen if you intend to continue working on it. Thank you!