Qwen4ExpForConditionalGeneration

Model type	Qwen4ExpForConditionalGeneration
Class	Qwen4ExpForConditionalGeneration
Task	qwen4-exp-vision-language
Source	models/qwen4_exp.py

Description

Qwen3.8 Flash-Next/Qwen4-Exp three-model vision-language pipeline.

Reuses the source-identical Qwen3/Qwen3.5 packed vision encoder and embedding mixer, while preserving the Qwen4-Exp decoder’s independent PLE token stream and heterogeneous recurrent, QSA, and position state.

Usage

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")