neo_chat¶


Model type	`neo_chat`
Class	`SenseNovaU1Model`
Task	`sensenova-u1`
Source	`models/sensenova_u1.py`

Description¶

SenseNova-U1.5 NEO-unify unified understanding + generation model.

Builds five ONNX graphs that mirror the upstream inference stages:

decoder Understanding decoder — text generation and VLM understanding. vision_encoder Reference-image tower feeding the understanding branch. embedding Token embedding with reference-image feature scatter. image_gen_embedding Noisy-latent tower + timestep / noise-scale embedders. image_gen_denoiser Generation-branch transformer + pixel-shuffle flow-matching head.

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")