Synthetic training data that teaches a small model (e.g.
google/functiongemma-270m-it) to map natural-language shopping requests
onto whatever tool schema is declared in the prompt — not onto one
hard-coded API.
A visitor chats with the store's corner chatbot: "I need an inexpensive
top-loading washing machine, preferably from a German manufacturer" → the
model must emit the declared tool call with the right query… See the full description on the dataset page:
https://huggingface.co/datasets/Qrzysztof/ecommerce-chat-tool-calling.
