Update app.py

app.py

@@ -14,7 +14,6 @@ torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
 pipe = pipe.to(device)
 
-# Maximum values as defined in your original code
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 1024
 
@@ -39,7 +38,6 @@ def infer(
     num_inference_steps=40,
     progress=gr.Progress(track_tqdm=True),
 ):
-    # Optionally randomize seed
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
     generator = torch.Generator(device=device).manual_seed(seed)
@@ -48,7 +46,10 @@ def infer(
     prompt = truncate_text(prompt, max_tokens=77)
     negative_prompt = truncate_text(negative_prompt, max_tokens=77) if negative_prompt.strip() else ""
 
-    # Explicitly set pad_token_id to eos_token_id for open-end generation.
+    # Move model to GPU before inference.
+    pipe.model.to("cuda")
+    
+    # Generate image using the truncated prompts.
     image = pipe(
         prompt=prompt,
         negative_prompt=negative_prompt,
@@ -57,12 +58,14 @@ def infer(
         width=width,
         height=height,
         generator=generator,
-        pad_token_id=pipe.tokenizer.eos_token_id,
     ).images[0]
 
+    # Move model back to CPU after inference.
+    pipe.model.to("cpu")
+
     return image, seed
 
-# Example prompt for testing
+# UI layout remains unchanged.
 examples = [
     "A capybara wearing a suit holding a sign that reads Hello World",
 ]
@@ -151,5 +154,5 @@ with gr.Blocks(css=css) as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)