wan example

examples/dynamo/wan/wan.py

@@ -0,0 +1,50 @@
+import argparse
+import torch
+from diffusers import StableDiffusionPipeline
+
+import torch_migraphx
+
+torch._dynamo.reset()
+
+parser = argparse.ArgumentParser(description='Conversion parameters')
+
+parser.add_argument('--fp16',
+                    action='store_true',
+                    help='Load fp16 version of the pipeline')
+
+
+def run(args):
+    from diffusers import FlowMatchEulerDiscreteScheduler, UniPCMultistepScheduler, WanPipeline
+    from diffusers.utils import export_to_video
+    # pip install git+https://github.com/huggingface/diffusers.git
+    # pip install ftfy
+
+    #scheduler_a = FlowMatchEulerDiscreteScheduler(shift=5.0)
+    #scheduler_b = UniPCMultistepScheduler(prediction_type="flow_prediction", use_flow_sigmas=True, flow_shift=4.0)
+
+    pipe = WanPipeline.from_pretrained("Wan-AI/Wan2.1-T2V-1.3B-Diffusers")
+
+    pipe = pipe.to("cuda")
+
+    pipe.text_encoder = torch.compile(pipe.text_encoder, backend='migraphx')
+    pipe.transformer = torch.compile(pipe.transformer, backend='migraphx')
+    pipe.vae.decoder = torch.compile(pipe.vae.decoder, backend='migraphx')
+
+    prompt = "A cat walks on the grass, realistic"
+    negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"
+
+    output = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        height=480,
+        width=832,
+        num_frames=81,
+        guidance_scale=5.0,
+    ).frames[0]
+    export_to_video(output, "output.mp4", fps=15)
+
+
+if __name__ == '__main__':
+    args = parser.parse_args()
+
+    run(args)