b10227

ggml-org/llama.cppb10227Aug 2, 2026by github-actions[bot]same day as b10225

What changed since b10225

medium risk

Adds a dedicated Qwen3 chat parser with tagged thinking and tool-call handling, plus chat parsing refactors.

New features

  • Qwen3 models now get a specialized chat parser instead of relying on the generic template handling.
  • New tagged thinking tool parser separates reasoning output from tool invocations for Qwen3-style models.
  • Chat context handling now supports omission of the <tool_call> marker when parsing model output.

Improvements

  • Chat parsing utilities were refactored and gained permute helper functions for cleaner message transformation.
  • Tool delimiters were updated and the <function trigger pattern was corrected for more reliable tool detection.
  • Added documentation comments covering Qwen3-Coder specifics.

Bug fixes

  • Fixed the trigger pattern for <function so tool calls are detected correctly in Qwen3 output.

Upgrade notes

This is a chat-layer change only; no config or model file migration is needed. If you serve Qwen3 or Qwen3-Coder models, re-test tool-calling and reasoning output since delimiters and omission handling changed. Other model families should behave as before.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

A new release adding specialized parsing support for Qwen3 models, including a tagged thinking tool parser and refactoring of chat handling utilities.

Key Highlights

  • Added Qwen3 specialized parser with tagged thinking tool support.
  • Refactored chat parsing and added permute helper functions.
  • Added support for omission handling in chat context.

Release Notes (excerpt)

<details open> chat : add qwen3 specialized parser (#26252) * Add tagged thinking tool parser * chat : refactor and add permute helper * cont : add support for <tool_call> omission * cont : update tool delimiters * cont : add comment for qwen3-coder * cont : fix trigger pattern for <function --------- Co-authored-by: Bart de Boer <bart.deboer@gmail.com> </details> **Website:** - <https://llama.app> **macOS/iOS:** - macOS Apple Silicon (arm64)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories