Update pipeline.py

pipeline.py

@@ -1496,17 +1496,6 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     start_guidance_time = time.time()
                         
                     if do_classifier_free_guidance:
-                        # # Split tensor along its first dimension
-                        # noise_pred_uncond, noise_pred_text = torch.chunk(noise_pred, 2, dim=0)
-
-                        # # Efficient in-place addition using advanced indexing
-                        # noise_pred_uncond_sum[..., current_context_indexes, :, :] += noise_pred_uncond
-                        # noise_pred_text_sum[..., current_context_indexes, :, :] += noise_pred_text
-
-                        # # Efficient in-place increment for latent_counter
-                        # latent_counter[current_context_indexes] += 1
-                        
-                        
                         # Split tensor along its first dimension
                         noise_pred_uncond, noise_pred_text = torch.chunk(noise_pred, 2, dim=0)
 
@@ -1518,10 +1507,18 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                         # Perform batch addition
                         noise_pred_uncond_sum[..., current_context_indexes, :, :] += expanded_noise_pred_uncond
                         noise_pred_text_sum[..., current_context_indexes, :, :] += expanded_noise_pred_text
+                        
+                        # print devices and shapes for everything
+                        print("noise_pred_uncond_sum", noise_pred_uncond_sum.device, noise_pred_uncond_sum.shape)
+                        print("noise_pred_text_sum", noise_pred_text_sum.device, noise_pred_text_sum.shape)
+                        print("expanded_noise_pred_uncond", expanded_noise_pred_uncond.device, expanded_noise_pred_uncond.shape)
+                        print("expanded_noise_pred_text", expanded_noise_pred_text.device, expanded_noise_pred_text.shape)
+                        print("current_context_latents", current_context_latents.device, current_context_latents.shape)
+                        print("current_context_indexes", current_context_indexes.device, current_context_indexes.shape)
+                        print("latent_counter", latent_counter.device, latent_counter.shape)
+                        
 
                         # Batch increment for latent_counter
-                        # Here, you need to ensure that the addition is done correctly depending on how latent_counter is structured
-                        # If latent_counter is a tensor, you might use something like:
                         latent_counter[current_context_indexes] += 1
                         
                     print("guidance time", time.time() - start_guidance_time)