Pin unslothai/llama.cpp#235 (keep unparsed Qwen3-Coder output as content)

[danielhanchen]

Adds #235 to the mix, last in the list.

#235 fixes the Qwen3-Coder parser (Qwen3.5 / 3.6 / 3.8, Nemotron Nano 3, StepFun) silently dropping any output after a <tool_call> that does not parse, such as a call to a tool that is not offered or a call cut off by EOS. The client got a reply that ended mid-sentence with finish_reason: "stop" and no tool call. In Unsloth Studio that showed up as a reply cut in the middle of a code block with no Continue button. Details and before / after tables are in #235.

Checks

• Composed locally the way the nightly does it (b11160, each pin in order with diff3 and the additive merge fallback, then chat : keep unparsed Qwen3-Coder output as content instead of dropping it #235): every pin merges, chat : keep unparsed Qwen3-Coder output as content instead of dropping it #235 (now at 6a5eca1) merges cleanly, and pin_contract.py reports all 14 pins intact.
• Recorded in feature-checks.json under unchecked: the change is a chat parser fix covered by test-chat, and none of the arch / backend-op / mtmd probes applies.
• The resolver-tests job passes locally (all eight resolver tests, the pin coverage lint and the workflow checks).
• chat : keep unparsed Qwen3-Coder output as content instead of dropping it #235 touches only common/chat.cpp, common/parsers/qwen3-coder.cpp and tests/test-chat.cpp. The only other pin touching common/chat.cpp is ggml-org#25731 (Inkling), and it composes with it.

Unrelated, found while checking

On newer base tags the existing pin set no longer merges (tonight's base is b11226): on b11205, ggml-org#24423 conflicts in src/llama-context.cpp and #176 in tests/test-mtmd-impl.cpp, before #235 is reached. The last three nightlies (09-25 to 09-27) failed at Resolve tag on exactly this: ggml-org/llama.cpp#24423 (3dac51dc8) does not merge cleanly onto b11222, so the last mix release is still b11160. This PR does not change that, and #235 only reaches a release once ggml-org#24423 is repinned.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-28T11:08:30.599066Z	ef7080a	PR opened
🔒 Security Review	✅ Completed 2026-09-28T11:10:32.658278Z	ef7080a	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ef7080a2ac

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Register the new pin in feature-checks

When this pr-set.json change is pushed or submitted in a PR, the Every pin is either checked or knowingly unchecked step in .github/workflows/unsloth-pr-set-lint.yml derives unslothai#235 from this URL and exits with status 1 because that key appears in neither features nor unchecked in scripts/unsloth/feature-checks.json. Add a feature check or a documented unchecked entry for this pin so the required lint workflow can pass.

Useful? React with 👍 / 👎.