Hi, what about LFM models?

[AndreyGasik]

Its not a issue, just a question, new model LFM2.5 is very good and fast, but they have some special instuctions for tools and think, now i use my own Python chat for use this model, can you realize this model work inside your chat? will be very cool. Thanks for your work

[yoloyash]

Hey Andrey, thanks for bringing this up. I am a huge fan of LiquidAI's work. LFM2.5 models should work out of the box for chat + thinking - we alrady render traces. LFM2 uses Pythonic function calls so on vLLM you'd serve with a --enable-auto-tool-choice --tool-call-parser pythonic (the model's own Jinja template handles the rest) and our web tools map straight through the OpenAI API. llama.cpp similar idea. If you're on something else, lmk what you're hosting with and we can figure it out.