Sign In

CausVid Tests (Wan2.1 1.3B T2V, Python 3.12, PyTorch 2.7, Cuda 12.8, SageAttention, AutoRegressive, RTX3060 12GB VRAM + 64GB RAM)

Loading Images

Using https://github.com/tianweiy/CausVid

1) Clone https://github.com/tianweiy/CausVid and follow instructions to install requirements

2) Clone https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B into wan_models/Wan2.1-T2V-1.3B

3) Put the pt file inside checkpoint_folder/model.pt

4) Run inference code, python minimal_inference/autoregressive_inference.py --config_path configs/wan_causal_dmd.yaml --checkpoint_folder XXX --output_folder XXX --prompt_file_path XXX

--

Also check out

Prebuilt Wheels (flash attention, sage attention, xformers): https://huggingface.co/datasets/siraxe/PrecompiledWheels_Torch-2.8-cu128-cp312/tree/main

Fork that supports 14B model: https://github.com/GoatWu/CausVid-Plus

My experimental fork using Sage Attention (if flash attention doesn't work, ymmv): https://github.com/YewMeng/CausVid/tree/sage_attention

Comments