Datasetpublic

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.

sourceHugging Faceupdated 8mo agoView on Hugging Face

1likes15downloads

Dataset Card

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.

—For HF Dataset Viewer and analysis only
—Not used for eval logic

Key fields

—conversation_id
—turn_index
—user
—expect_search_history

temporal

Evaluates temporal parsing correctness when invoking history search.

—Ground-truth start/end timestamps
—Sanity check for search term extraction
—Does not evaluate tool invocation or retrieval quality

Load with pandas

python

from datasets import load_dataset

# intent (conversation-level)
intent = load_dataset(
    "Mozilla/history-search-tool-calling",
    "intent",
    revision="v1.0.0",
)["train"].to_pandas()

# intent_turns (turn-level)
intent_turns = load_dataset(
    "Mozilla/history-search-tool-calling",
    "intent_turns",
    revision="v1.0.0",
)["train"].to_pandas()

# temporal
temporal = load_dataset(
    "Mozilla/history-search-tool-calling",
    "temporal",
    revision="v1.0.0",
)["train"].to_pandas()