Optimization Examples

Wan 2.1 QAT Attention 14B Inference

Use these files for Wan 2.1 14B inference with the ATTN_QAT_INFER backend:

examples/inference/optimizations/download_14B_qat.sh
examples/inference/optimizations/attn_qat_inference_example.py

1. Download the 14B QAT checkpoint

The helper script downloads the QAT safetensors from FastVideo/14B_qat_400 into checkpoints/14B_qat_400 by default.

Prerequisites:

huggingface_hub installed, for example: uv pip install huggingface_hub
access to the model repo if it is private or gated: huggingface-cli login

Run:

bash examples/inference/optimizations/download_14B_qat.sh

To download into a custom directory, pass it as the first argument:

bash examples/inference/optimizations/download_14B_qat.sh /path/to/14B_qat_400

2. Edit the inference example for Wan 2.1 14B

Open examples/inference/optimizations/attn_qat_inference_example.py and update these two values:

Change the base model from Wan-AI/Wan2.1-T2V-1.3B-Diffusers to Wan-AI/Wan2.1-T2V-14B-Diffusers.
Replace the placeholder init_weights_from_safetensors="safetensors_path" with the directory that contains the downloaded .safetensors files.

Example:

generator = VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-14B-Diffusers",
    num_gpus=1,
    use_fsdp_inference=True,
    dit_cpu_offload=False,
    vae_cpu_offload=False,
    text_encoder_cpu_offload=True,
    pin_cpu_memory=False,
    init_weights_from_safetensors="checkpoints/14B_qat_400",
)

The script already sets:

os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "ATTN_QAT_INFER"

3. Run the example

python examples/inference/optimizations/attn_qat_inference_example.py

The generated videos are written to video_samples/ by default.

Notes

ATTN_QAT_INFER requires the in-repo fastvideo-kernel build to expose the attn_qat_infer package.
If you have not built the kernel yet, run cd fastvideo-kernel && ./build.sh first.
If you keep the example on the 1.3B base model while loading the 14B QAT weights, the model/config will not match.