Insight-V-Reason-LLaMA3

Model Summary

The Insight-V models are 7B parameter models based on LLaMA3-8B language model with a context window of 32K tokens.

Insight-V offers 1) a scalable data generation pipeline for long-chain, high-quality reasoning data, 2) a multi-agent system that decomposes visual reasoning tasks into reasoning and summarization, and 3) a two-stage training pipeline to enhance visual reasoning capabilities. Together, these contributions address key challenges in visual reasoning, providing a solid foundation for future research in MLLM reasoning.

• Repository: https://github.com/dongyh20/Insight-V
• Languages: English, Chinese
• Paper: https://arxiv.org/abs/2411.14432

Model Architecture

• Architecture: Pre-trained Oryx-ViT + LLaMA3-8B
• Data: a mixture of 200k reasoning data
• Precision: BFloat16

Hardware & Software

• Hardware: 64 * NVIDIA Tesla A100
• Orchestration: HuggingFace Trainer
• Code: Pytorch

Citation