Omnivision-968M: Vision Language Model with 9x Tokens Reduction for Edge Devices

OmniVision-968M is heralded as the world's smallest vision language model optimized for edge devices, providing a 9x token reduction over traditional models, achieving enhanced accuracy and reduced computational costs.

• Sub-billion model
• Enhances accuracy
• 9x token reduction
• Uses DPO training
• Optimized for edge