Update app.py

app.py

@@ -96,7 +96,7 @@ class ModelWrapper:
             current_timesteps = torch.ones(len(prompt_embed), device="cuda", dtype=torch.long) * constant
             current_timesteps = current_timesteps.to(torch.float16)
             print(f'current_timestpes: {current_timesteps.dtype}')
-            eval_images = self.model(noise.to(torch.float16), current_timesteps, prompt_embed.to(torch.float16), added_cond_kwargs=unet_added_conditions).sample
+            eval_images = self.model(noise, current_timesteps, prompt_embed, added_cond_kwargs=unet_added_conditions).sample
             print(type(eval_images))
 
             eval_images = get_x0_from_noise(noise, eval_images, alphas_cumprod, current_timesteps).to(self.DTYPE)
@@ -123,7 +123,7 @@ class ModelWrapper:
 
         add_time_ids = self.build_condition_input(height, width).repeat(num_images, 1)
 
-        noise = torch.randn(num_images, 4, height // self.vae_downsample_ratio, width // self.vae_downsample_ratio, generator=generator).to(device="cuda", dtype=torch.float16) 
+        noise = torch.randn(num_images, 4, height // self.vae_downsample_ratio, width // self.vae_downsample_ratio, generator=generator).to(device="cuda").to(torch.float16)
 
         prompt_inputs = self._encode_prompt(prompt)
         
@@ -132,13 +132,13 @@ class ModelWrapper:
         prompt_embeds, pooled_prompt_embeds = self.text_encoder(prompt_inputs)
 
         batch_prompt_embeds, batch_pooled_prompt_embeds = (
-            prompt_embeds.repeat(num_images, 1, 1),
-            pooled_prompt_embeds.repeat(num_images, 1, 1)
+            prompt_embeds.repeat(num_images, 1, 1).to(torch.float16),
+            pooled_prompt_embeds.repeat(num_images, 1, 1).to(torch.float16)
         )
 
         unet_added_conditions = {
             "time_ids": add_time_ids,
-            "text_embeds": batch_pooled_prompt_embeds.squeeze(1).to(torch.float16)
+            "text_embeds": batch_pooled_prompt_embeds.squeeze(1)
         }