qwen2_vl_text¶


Model type	`qwen2_vl_text`
Class	`Qwen25VLTextModel`
Task	`text-generation`
Source	`models/qwen_vl.py`

Description¶

Qwen2.5-VL text-only decoder.

Extracts the text backbone from the Qwen2.5-VL multimodal model. Strips language_model. weight prefixes and drops visual. keys. For text-only inference the standard 1D RoPE is equivalent to MRoPE (all three dimensions are identical for text tokens).

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")