Update pipeline.py

pipeline.py

@@ -972,9 +972,6 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     if context_group == 0:current_context_start = 0
                     else:current_context_start = context_group * (context_size - overlap)
 
-                    # print current start, total frames, and context size, and if end frame is greater than total frames
-                    print(f"Current context start: {current_context_start}, total frames: {num_frames}, context size: {context_size}, end frame: {current_context_start + context_size}")
-                    
                     # select the relevent context from the latents
                     current_context_latents = latents[:, :, current_context_start : current_context_start + context_size, :, :]
                     
@@ -982,11 +979,7 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     
                     # if context_start + context_size > num_frames: append the remaining frames from the start of the latents
                     if wrap_count > 0:
-                        print(f"Appending {wrap_count} frames from the start of the latents")    
                         current_context_latents = torch.cat([current_context_latents, latents[:, :, :wrap_count, :, :]], dim=2)
-                        
-                    # print number of frames in the context
-                    print(f"Number of frames in the context: {current_context_latents.shape[2]}")
                 
                     # expand the latents if we are doing classifier free guidance
                     latent_model_input = torch.cat([current_context_latents] * 2) if do_classifier_free_guidance else current_context_latents
@@ -1012,6 +1005,9 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                         current_context_latents = current_context_latents[:, :, :-wrap_count, :, :]
                         # remove the ending frames from noise_pred
                         noise_pred = noise_pred[:, :, :-wrap_count, :, :]
+                        # print the shape of the current_context_latents and noise_pred
+                        print(f"current_context_latents shape: {current_context_latents.shape}")
+                        print(f"noise_pred shape: {noise_pred.shape}")
                     
                     # compute the previous noisy sample x_t -> x_t-1
                     current_context_latents = self.scheduler.step(noise_pred, t, current_context_latents, **extra_step_kwargs).prev_sample