phi-3-vision-onnx

phi-3-vision-onnx is an ONNX int4 quantized version of microsoft/Phi-3-vision-128k-instruct, providing an inference implementation, optimized for AI PCs.

This is a vision-to-text model from the Phi3 release series and is a very high-quality innovative small model that accepts multi-modal inputs (image/video, text).

Model Description

• Developed by: microsoft
• Quantized by: microsoft
• Model type: phi-3-vision
• Parameters: 3.8 billion
• Model Parent: microsoft/Phi-3-vision-128k-instruct
• Language(s) (NLP): English
• License: Apache 2.0
• Uses: Multimodal LLM
• Quantization: int4

Model Card Contact

llmware on github

llmware on hf

llmware website