Skip to content

Dataset · xiuhuywh

Cold-start SFT for image reasoning

xiuhuywh/DRIM-ColdstartSFT

Cold-start dataset for SFT models learning multi-turn reasoning with images.

What's inside

Contains cold-start SFT data from the paper Deep But Reliable: Advancing Multi-turn Reasoning for Thinking with Images. Modalities are image and text, size is 10K–100K. Used in the multi-turn reasoning research for Thinking with Images.

What you can do with it

  • SFT vision-language models for multi-turn image reasoning
  • LoRA adapter for VLM
  • Testing the reasoning with images training pipeline
Size
10K–100K rows
Downloads this month
566
Likes
117

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("xiuhuywh/DRIM-ColdstartSFT")