Enable the i2vgen pipeline

examples/stable-diffusion/i2v.py

@@ -0,0 +1,39 @@
+import torch
+from diffusers.utils import export_to_gif, load_image
+
+from optimum.habana.diffusers import GaudiI2VGenXLPipeline
+
+def main() :
+    torch._C._set_math_sdp_allow_fp16_bf16_reduction(True)
+    pipeline = GaudiI2VGenXLPipeline.from_pretrained(
+        "ali-vilab/i2vgen-xl",
+        torch_dtype=torch.bfloat16,
+        use_hpu_graphs=False,
+        use_habana=True,
+        gaudi_config="Habana/stable-diffusion",
+    )
+    #torch_dtype=torch.float,
+    # import pdb;pdb.set_trace()
+    # breakpoint()
+    pipeline.enable_model_cpu_offload(device="hpu")
+
+    image_url = "https://huggingface.co/datasets/diffusers/docs-images/resolve/main/i2vgen_xl_images/img_0009.png"
+    image = load_image(image_url).convert("RGB")
+
+    prompt = "Papers were floating in the air on a table in the library"
+    negative_prompt = "Distorted, discontinuous, Ugly, blurry, low resolution, motionless, static, disfigured, disconnected limbs, Ugly faces, incomplete arms"
+    generator = torch.manual_seed(8888)
+    # import pdb;pdb.set_trace()
+    # breakpoint()
+    frames = pipeline(
+        prompt=prompt,
+        image=image,
+        num_inference_steps=50,
+        negative_prompt=negative_prompt,
+        guidance_scale=9.0,
+        generator=generator,
+    ).frames[0]
+    export_to_gif(frames, "i2v.gif")
+
+if __name__ == "__main__":
+    main()

optimum/habana/diffusers/__init__.py

@@ -4,6 +4,7 @@
     GaudiStableVideoDiffusionControlNetPipeline,
 )
 from .pipelines.ddpm.pipeline_ddpm import GaudiDDPMPipeline
+from .pipelines.i2vgen_xl.pipeline_i2vgen_xl import GaudiI2VGenXLPipeline
 from .pipelines.flux.pipeline_flux import GaudiFluxPipeline
 from .pipelines.flux.pipeline_flux_img2img import GaudiFluxImg2ImgPipeline
 from .pipelines.pipeline_utils import GaudiDiffusionPipeline

optimum/habana/diffusers/models/unet_2d_condition.py

@@ -100,7 +100,7 @@ def gaudi_unet_2d_condition_model_forward(
     timesteps = timesteps.expand(sample.shape[0])
 
     t_emb = self.time_proj(timesteps)
-
+    print(f"mark_step")
     import habana_frameworks.torch.core as htcore
 
     htcore.mark_step()