Update pipeline.py

pipeline.py

@@ -1058,6 +1058,10 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                             noise_pred_text = noise_pred_text[:, :, :-wrap_count, :, :]
                         noise_pred_uncond_sum[:, :, current_context_start : current_context_start + context_size, :, :] += noise_pred_uncond
                         noise_pred_text_sum[:, :, current_context_start : current_context_start + context_size, :, :] += noise_pred_text
+                        
+                        # print min and max values of noise_pred_uncond_sum and noise_pred_text_sum
+                        print(f"noise_pred_uncond_sum min: {noise_pred_uncond_sum.min()} max: {noise_pred_uncond_sum.max()}")
+                        print(f"noise_pred_text_sum min: {noise_pred_text_sum.min()} max: {noise_pred_text_sum.max()}")
 
                     # set the step index to the current batch
                     self.scheduler._step_index = i
@@ -1065,12 +1069,14 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                 # perform guidance
                 if do_classifier_free_guidance:
                     latent_counter = latent_counter.reshape(1, 1, num_frames, 1, 1)
+                    print(f"latent_counter min: {latent_counter.min()} max: {latent_counter.max()}")
                     noise_pred_uncond = noise_pred_uncond_sum / latent_counter
                     noise_pred_text = noise_pred_text_sum / latent_counter
                     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
-                
+                    
                 # compute the previous noisy sample x_t -> x_t-1
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+                print(f"latents min: {latents.min()} max: {latents.max()}")
                 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):