update

inference.py

@@ -354,6 +354,7 @@ def inference(network: LoRANetwork, tokenizer: CLIPTokenizer, text_encoder: CLIP
                 latent_model_input = noise_scheduler.scale_model_input(latent_model_input, timestep=t).to(weight_dtype)
                 # predict the noise residual
                 with network:
+                    print(f"dtype: {latent_model_input.dtype}, {text_embedding.dtype}, t={t}")
                     noise_pred = unet(latent_model_input, t , encoder_hidden_states=text_embedding).sample
 
                 # perform guidance
@@ -373,7 +374,7 @@ def inference(network: LoRANetwork, tokenizer: CLIPTokenizer, text_encoder: CLIP
             with torch.no_grad():
                 image = vae.decode(latents).sample
             image = (image / 2 + 0.5).clamp(0, 1)
-            image = image.detach().cpu().permute(0, 2, 3, 1).numpy()
+            image = image.detach().cpu().permute(0, 2, 3, 1).to(torch.float32).numpy()
             images = (image * 255).round().astype("uint8")