Info: llama.cpp-related issue when using docker-model-runner with gpt-oss and tooling enabled

[optimisticupdate]

While testing docker-model-runner with gpt-oss, I noticed a problem that only occurs when tooling is enabled.
This is not an issue with docker-model-runner itself, it appears to be caused by upstream changes in llama.cpp.

For reference, there’s an open PR in llama.cpp that may resolve it:
ggml-org/llama.cpp#15158

Just sharing this here so others are aware, and looking forward to a new release image once the fix is available upstream.

Since I was not that active in open source contribution the last years I hope it is fine to post it here.

[kiview]

Thanks you @optimisticupdate, it is absolutely fine to post here and sharing the info and the link to the upstream PR is much appreciated.

[optimisticupdate]

There was a merge to main which should fix/improve usage of llama.cpp, I am going to check that out. Just letting you know.

ggml-org/llama.cpp#15181

[kiview]

This should be fixed by ggml-org/llama.cpp#15181, I guess @p1-0tr will roll out the llama.cpp backend update by next week.

[xenoscopic]

It looks like we're now up-to-date with b6191 (and the fix is available from b6173 onward), so I think we can close this. Note that you'll need to restart Docker Desktop to pick up the update (or, if using Docker CE, do docker model uninstall-runner --images && docker model install-runner).

Feel free to re-open if problem persists.