2,101 image-text pairs designed for unsupervised post-training of multi-modal large language models. Each entry includes a 'problem' field with a geometric reasoning question and an 'answer' field containing the corresponding solution.
Use Cases
- Train multi-modal LLMs using the 'problem' and 'answer' fields to improve geometric reasoning capabilities.
- Implement Group Relative Policy Optimization (GRPO) by utilizing the synthesized reasoning paths.
- Fine-tune vision-language models for unsupervised post-training scenarios using the image-text pairs.
Strengths
- 2,101 examples of geometric reasoning problems paired with textual solutions.
- Includes a 'problem' column containing multi-modal questions requiring spatial reasoning.
- Features an 'answer' column providing step-by-step solutions for model supervision.
- Derived from the Geometry3K benchmark to support the MM-UPT post-training framework.