CausalForcing-Wan2.1-T2V-1.3B-Diffusers

Diffusers-layout conversion of the chunk-wise Causal Forcing generator (zhuhz22/Causal-Forcing, chunkwise/causal_forcing.pt, generator weights) from thu-ml/Causal-Forcing — "Causal Forcing: Autoregressive Diffusion Distillation Done Right" (ICML 2026).

Wan2.1-T2V-1.3B backbone, block-causal DMD student: 3-latent-frame chunks, 21-latent-frame sliding attention window, 4 warped denoising steps (1000/750/500/250, shift 5.0), no CFG. Non-transformer components are copied from Wan-AI/Wan2.1-T2V-1.3B-Diffusers.

Converted with sglang.multimodal_gen.tools.convert_forcing_to_diffusers --preset causal-forcing-chunkwise for use with the SGLang diffusion runtime:

sglang generate --model-path frankleeeee/CausalForcing-Wan2.1-T2V-1.3B-Diffusers \
  --prompt "A stylish woman walks down a Tokyo street..." \
  --width 832 --height 480 --num-frames 81 --save-output

Note: the upstream project releases only Wan 1.3B generators; the Wan2.1-T2V-14B named in its configs is the DMD teacher (real_name), not a released generator.