Tool parser + reasoning parser?
Please include these types of configs for vLLM/SGLang/etc. in the model card like the rest of the industry.
Thanks for your suggestion!
We've just updated the model cards of the Luciole-Instruct-1.1 and Luciole-Thinking-1.1 models with the vLLM configs for tool calling (and the reasoning parser for the Thinking models). See for instance the tool calling section of Luciole-23B-Instruct-1.1.
Luciole-23B-Base itself is a pretrained base model with no chat, tool or reasoning post-training, so no parser applies to it. Please use the Instruct or Thinking variants for that.
Of course, and thank you for including the info.
Will you release FP8 for thinking models?
Thank you for your suggestions @pathosethoslogos
We added FP8 versions of thinking models in our collections:
https://huggingface.co/OpenLLM-France/Luciole-23B-Thinking-1.1-FP8
https://huggingface.co/OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1-FP8
Ah, I see it in the collection now. Thank you!