wlsaidhi commited on
Commit
e9db6c5
·
verified ·
1 Parent(s): a6e1294

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +47 -0
README.md CHANGED
@@ -35,3 +35,50 @@ We're excited to introduce the **CausalWan2.2 I2V A14B series**—a new line of
35
 
36
  - 8-step inference is supported.
37
  - Try it out on **FastVideo** — we support a wide range of GPUs from **H100** to **4090**, and also support **Mac** users!
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35
 
36
  - 8-step inference is supported.
37
  - Try it out on **FastVideo** — we support a wide range of GPUs from **H100** to **4090**, and also support **Mac** users!
38
+
39
+ ## Inference code
40
+
41
+ ```python
42
+ from fastvideo import VideoGenerator, SamplingParam
43
+ import json
44
+ # from fastvideo.configs.sample import SamplingParam
45
+
46
+ OUTPUT_PATH = "video_samples_self_forcing_causal_wan2_2_14B_i2v"
47
+ def main():
48
+ # FastVideo will automatically use the optimal default arguments for the
49
+ # model.
50
+ # If a local path is provided, FastVideo will make a best effort
51
+ # attempt to identify the optimal arguments.
52
+ generator = VideoGenerator.from_pretrained(
53
+ "FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers",
54
+ # FastVideo will automatically handle distributed setup
55
+ num_gpus=1,
56
+ use_fsdp_inference=True,
57
+ dit_cpu_offload=True, # DiT need to be offloaded for MoE
58
+ dit_precision="fp32",
59
+ vae_cpu_offload=False,
60
+ text_encoder_cpu_offload=True,
61
+ dmd_denoising_steps=[1000, 850, 700, 550, 350, 275, 200, 125],
62
+ # Set pin_cpu_memory to false if CPU RAM is limited and there're no frequent CPU-GPU transfer
63
+ pin_cpu_memory=True,
64
+ # image_encoder_cpu_offload=False,
65
+ )
66
+
67
+ sampling_param = SamplingParam.from_pretrained("FastVideo/SFWan2.2-I2V-A14B-Preview-Diffusers")
68
+ sampling_param.num_frames = 73
69
+ sampling_param.width = 832
70
+ sampling_param.height = 480
71
+ sampling_param.seed = 1000
72
+
73
+ with open("prompts/mixkit_i2v.jsonl", "r") as f:
74
+ prompt_image_pairs = json.load(f)
75
+
76
+ for prompt_image_pair in prompt_image_pairs:
77
+ prompt = prompt_image_pair["prompt"]
78
+ image_path = prompt_image_pair["image_path"]
79
+ _ = generator.generate_video(prompt, image_path=image_path, output_path=OUTPUT_PATH, save_video=True, sampling_param=sampling_param)
80
+
81
+
82
+ if __name__ == "__main__":
83
+ main()
84
+ ```