Datasetpublic

ali5341/scifact-chat-format

SciFact (Chat-Format Preparation) This dataset is a chat-format preparation of SciFact for supervised fine-tuning (SFT). Format This format is commonly referred to as: chat-format SFT data instruction-tuning conversations OpenAI-style messages format Included files train.jsonl validation.jsonl stats.json prepare_scifact_unsloth.py Source Base dataset: allenai/scifact Original Dataset Highlights Original dataset:… See the full description on the dataset page: https://huggingface.co/datasets/ali5341/scifact-chat-format.

sourceHugging Facecc-by-nc-2.0updated 5mo agoView on Hugging Face

0likes37downloads

Dataset Card

SciFact (Chat-Format Preparation)

This dataset is a chat-format preparation of SciFact for supervised fine-tuning (SFT).

Format

This format is commonly referred to as:

—chat-format SFT data
—instruction-tuning conversations
—OpenAI-style messages format

Included files

—train.jsonl
—validation.jsonl
—stats.json
—prepare_scifact_unsloth.py

Source

—Base dataset: allenai/scifact

Original Dataset Highlights

—Original dataset: allenai/scifact
—Focus: scientific claim verification with rationales.
—Labels: SUPPORT, CONTRADICT, and NOT ENOUGH INFO style supervision.
—Reported scale on source card: ~1.4K claims with evidence-containing abstracts.
—Paper: Fact or Fiction: Verifying Scientific Claims

Preparation summary

—Claim verification labels:
—SUPPORTS
—CONTRADICTS
—NOT ENOUGH INFORMATION
—Abstract sentences are indexed ([0], [1], ...).
—One row is emitted per claim-evidence set.
—NEI rows are created when no support/contradict evidence is present.

Assistant response format:

—Label: ...
—Rationale sentence ids: [...]
—Explanation: ...

Schema

Each JSONL row contains:

—messages
—user: instruction + claim + title + indexed abstract
—assistant: structured label/rationale/explanation
—meta: claim/doc ids, label, rationale sentence ids, split, variant

Reproduction

bash

python prepare_scifact_unsloth.py