GGUF
English
qwen3
reasoning
distillation
claude-opus
full-finetune
conversational

Create qwen3-think.jinja file with fixed tool calling

#1

I was getting wrong tool calling with the chat template suggested in the README:

<tool_group>
<tool_call>
<name=glob</name> 
<parameter=pattern> */ </parameter> 
</function> 
</tool_call> 
<tool_call> 
<name=read</name> 
<parameter=filePath> /path/to/file </parameter> 
</function> 
</tool_call> 
<tool_call> 
<name=read</name> 
<parameter=filePath> /path/to/file </parameter> 
</function> 
</tool_call>

This new chat template seems to fix the issue, I've been running it in OpenCode.

Thanks for the contribution! The simplified template in the README was dumping parameters as raw JSON instead of the proper XML format the model was pretrained on - good catch! Merging now!

samuelcardillo changed pull request status to merged

Sign up or log in to comment