Mozilla/history-search-tool-calling

History Search – Evaluation Datasets This repository contains evaluation-only datasets for history search tool usage. Datasets intent (canonical) Evaluates whether the model should invoke history search in a multi-turn conversation. One row per conversation Preserves full context Used by eval code Key fields id turns[] → { turn_index, user, expect_search_history } intent_turns (derived) Flattened, one row per turn view of intent.… See the full description on the dataset page: https://huggingface.co/datasets/Mozilla/history-search-tool-calling.