reasoning-degeneration-dev/

prepretraining-sft-v1

Tulu 3 SFT subset (50K examples) for pre-pretraining eval. Applied identically to all experimental conditions.

Rows: 50000 Columns: 1

	Columns




	

Column Type Description

messages List({'content': Value('string'), 'role': Value('string')}) Conversation turns [{role, content}] in Tulu 3 format

{ "script_name": "data/upload_data.py", "model": "N/A (SFT training data, not model… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/prepretraining-sft-v1.