standard

app.py

@@ -106,11 +106,7 @@ class ModelManager:
             self.single_model_name, self.device
         )
         set_default_chat_template(self.single_tokenizer, self.single_model_name)
-        
-        # Move to CUDA if available (following HuggingFace ZeroGPU pattern)
-        if torch.cuda.is_available():
-            self.single_model = self.single_model.to('cuda')
-        print(f"[Single] ✓ Model loaded")
+        print("[Single] ✓ Model loaded")
     
     def _load_t2t_model(self):
         """Load two-stage model."""
@@ -126,11 +122,7 @@ class ModelManager:
             device=str(self.device),
             background_prompt=self.t2t_background_prompt
         )
-        # Move to CUDA if available (following HuggingFace ZeroGPU pattern)
-        if torch.cuda.is_available():
-            self.t2t_model.context_model = self.t2t_model.context_model.to('cuda')
-            self.t2t_model.answer_model = self.t2t_model.answer_model.to('cuda')
-        print("[T2T] ✓ Models loaded")
+        print("[T2T] ✓ Model loaded")
     
     def _load_c2c_model(self):
         """Load Rosetta (C2C) model."""
@@ -187,10 +179,6 @@ class ModelManager:
         self.c2c_model, self.c2c_tokenizer = load_rosetta_model(
             model_config, eval_config, self.device
         )
-        
-        # Move to CUDA if available (following HuggingFace ZeroGPU pattern)
-        if torch.cuda.is_available():
-            self.c2c_model = self.c2c_model.to('cuda')
         print("[C2C] ✓ Model loaded")
     
     def _load_all_models(self):
@@ -261,15 +249,9 @@ class ModelManager:
         
         # Stream tokens
         generated_text = ""
-        try:
-            for token in streamer:
-                generated_text += token
-                yield generated_text
-        except Exception as e:
-            print(f"[Single] Streaming error: {e}")
-            yield f"Error generating response: {e}"
-        finally:
-            thread.join()
+        for token in streamer:
+            generated_text += token
+            yield generated_text
     
     @spaces.GPU(duration=90)
     def generate_t2t(self, user_input: str) -> Generator[tuple[str, str], None, None]:
@@ -399,15 +381,9 @@ class ModelManager:
         
         # Stream tokens
         generated_text = ""
-        try:
-            for token in streamer:
-                generated_text += token
-                yield generated_text
-        except Exception as e:
-            print(f"[C2C] Streaming error: {e}")
-            yield f"Error generating response: {e}"
-        finally:
-            thread.join()
+        for token in streamer:
+            generated_text += token
+            yield generated_text
 
 
 def create_demo(model_manager: ModelManager):
@@ -421,7 +397,12 @@ A. Why the act of destroying nature might be immoral.
 B. Why people who destroy the environment might be bad people.
 C. How the decision to preserve the environment benefits the environment.
 D. Whether plants have interests.""",
-        "example2": "Which company launched the Gemini 1.5 Pro model in early 2024?"
+        "example2": """Why is the Mars Exploration Rover Spirit currently tilted towards the north?
+
+A. Because it’s climbing up a big hill.
+B. Because it’s in the southern hemisphere where it is winter now.
+C. Because it’s in the northern hemisphere where it is winter now.
+D. Because one of its wheels broke."""
     }
     
     def respond(user_input: str):
@@ -488,7 +469,7 @@ D. Whether plants have interests.""",
         gr.Markdown("Example Questions:")
         with gr.Row():
             example1_btn = gr.Button("📝 Example 1: Philosophy", size="sm")
-            example2_btn = gr.Button("📝 Example 2: Knowledge Cutoff", size="sm")
+            example2_btn = gr.Button("📝 Example 2: Astronomy", size="sm")
 
 
         with gr.Row():