Instructions to use froggeric/Qwen-Fixed-Chat-Templates with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use froggeric/Qwen-Fixed-Chat-Templates with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Qwen-Fixed-Chat-Templates froggeric/Qwen-Fixed-Chat-Templates
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Github Copilot Desktop Canvas
This template works great thx :)
I'm encountering a weird error and I'm not sure it's in the harness or the template. We are using the github copilot app and the qwen3.8 struggles calling the canvas tool.
It claims to have it called like this
<function=...open_canvas>
<parameter name="canvasId">browser</parameter>
<parameter name="instanceId">browser-demo</parameter>
<parameter name="input">{"url": "https://example.com"}</parameter>
</function>
I'm not sure the serialization is correct in this case. Other tools so far worked flawlessly, and that is what confuses me right now.
Hey @AlexEBZJ , Copilot expects standard OpenAI JSON tool calls instead of Qwen's native XML format.
Try switching the template to JSON mode by passing tool_call_format: 'json' in your kwargs, or edit line 2 of chat_template.jinja to _tool_format = 'json'. That should get Copilot parsing the calls properly.
Thanks I'll try that som more. initial tests did not quit work (model was unabl to call any tools ), but this may be on me 😃. I wonder why all other copilot tools work except this one.
Ah saw that: "tool_call_format="json", use --tool-call-parser hermes." will do
