Keitsuna123/function-calling-demo

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

title: Small-Model Function Calling Demo emoji: 🔧 colorFrom: indigo colorTo: purple sdk: gradio sdkversion: 4.44.0 appfile: app.py pinned: false ---

Small-Model Function Calling — interactive demo

Try four fine-tuned Llama-3.2-1B models for function calling and compare how they handle relevant vs irrelevant queries. The headline: v1 SFT tends to always call a function (even nonsensically), while Path B (refusal-SFT) learns to refuse when no tool fits.

Companion to the full study: https://github.com/keitake123/llama-function-calling-study

Models: v1 · v2 · GRPO · Path B

Runs on free CPU — responses take ~15-30s; switching models reloads.