phi3_v

Model type

phi3_v

Class

Phi3VModel

Task

vision-language

Source

models/phi3_v.py

Description

Phi-3-Vision / Phi-3.5-Vision model (3-model split).

Builds three separate ONNX models:

  • decoder: Phi3 text decoder taking inputs_embeds

  • vision_encoder: CLIP ViT-L/14-336 + MLP projector

  • embedding: token embedding + image feature fusion

Supports microsoft/Phi-3-vision-128k-instruct and microsoft/Phi-3.5-vision-instruct.

Usage

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")