Runway· Gen / AlephResearch paper

Runway presents A2D-VL 7B, a diffusion vision language model adapted from the pretrained autoregressive VLM Qwen2.5-VL for parallel diffusion decoding. The model uses block size annealing and noise level annealing to preserve base model capabilities while enabling a speed-quality trade-off through adjustable parallelism.

From the source

We develop a state-of-the-art diffusion vision language model, Autoregressive-to-Diffusion (A2D), by adapting an existing autoregressive vision language model for parallel diffusion decoding.

runway.com

Announced