support gpt-oss function/reasoning in /v1/chat/completions#3962
support gpt-oss function/reasoning in /v1/chat/completions#3962lvhan028 merged 14 commits intoInternLM:mainfrom
Conversation
|
@mizuikk Thanks for your feedback.
This should be fixed.
Could you provide more detail information like inputs ? |
|
which parser should be specified? |
CMD: openwebui: roo code(can not use any tool): claude(In Claude, this model seems unable to understand human instructions.): Codex CLI seems to be just an official compatibility issue. |
|
Due to network issues, I can't log in to Codex. I tried the Roo-Code plugin and found that when calling the inference engine(lmdeploy/vllm), it don't pass tools params. Instead it instruct the LLM to output the result with a specific tag. If the specific tag is missing, Roo-Code assumes there’s an issue with the result (displaying 'Roo is having trouble'). I test vLLM and find its success rate to be higher. Comparing the inputs, I find two differences. First, according to this document, the system message should be treated as a developer message in gpt-oss. vLLM doesn’t handle this correctly, while LMDeploy does. The second difference is that for inputs without tools, vLLM removes the commentary channel. After removing this field, LMDeploy showed a significant improvement in success rate. (I'm not sure if this would also be helpful for the use of OpenWebUI)
I am not sure it is related to computational precision in rmsnorm, will test it soon. |
Is there a way to set the |
No, you can only set it in in request. According to your log, you actually set it. You can find I run your query ("Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option.") locally, but the response fall into repetition. We encountered this on some test datasets and suspected it is caused by rmsnorm but not yet tested |
|
Does the Offline Inference Pipeline not support this at all? @RunningLeon @irexyc |
|
@radna0 tool parser/reasoning parser only support serving api. |








Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily receiving feedbacks. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.
Motivation
Please describe the motivation of this PR and the goal you want to achieve through this PR.
Modification
Please briefly describe what modification is made in this PR.
BC-breaking (Optional)
Does the modification introduce changes that break the backward-compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.
Use cases (Optional)
If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.
Checklist