qwen2_5_vl

Model type	qwen2_5_vl
Class	Qwen25VLCausalLMModel
Task	qwen-vl
Source	models/qwen_vl.py

Description

Qwen2.5-VL vision-language model (3-model split).

Builds three separate ONNX models for onnxruntime-genai:

• decoder: text decoder taking inputs_embeds (MRoPE position_ids)

• vision_encoder: vision ViT with windowed/full attention

• embedding: token embedding + image feature fusion

The :class:~mobius.tasks.Qwen25VL3ModelTask calls each sub-module separately to produce 3 ONNX graphs.

Usage

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")