Update app.py

app.py

@@ -519,18 +519,7 @@ class SpeedGenerator(ARDiffusionGenerator):
         
         return response
 
-@spaces.GPU
-def load_model():
-    """Load model with Zero GPU optimization using @spaces.GPU"""
-    global tokenizer, model, device
-    
-    if tokenizer is not None and model is not None:
-        return tokenizer, model, device
-    
-    model_path = "rootxhacker/llama-3B-diffusion-exp-fixed"
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    
-    print(f"Loading model on {device}...")
+ {device}...")
     
     tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
     if tokenizer.pad_token is None:
@@ -631,6 +620,7 @@ def create_interface():
             <p><strong>⚠️ EXPERIMENTAL MODEL ⚠️</strong></p>
             <p>This is an experimental AR-Diffusion model. Results may vary and the model is still under development.</p>
             <p><em>🔥 Powered by Zero GPU with @spaces.GPU</em></p>
+            <p><small>Model: rootxhacker/llama-3B-diffusion-exp-fixed (LoRA Adapter)</small></p>
         </div>
         """)
         
@@ -682,7 +672,8 @@ def create_interface():
                     <h3>ℹ️ About AR-Diffusion</h3>
                     <p>This experimental model uses autoregressive diffusion for text generation, creating responses by iteratively denoising masked tokens.</p>
                     <br>
-                    <p><strong>Note:</strong> This model is experimental and may produce unexpected results.</p>
+                    <p><strong>Model:</strong> LoRA adapter trained for AR-Diffusion</p>
+                    <p><strong>Note:</strong> This model is experimental and may produce unexpected results. If the specific model fails to load, a fallback model will be used for demonstration.</p>
                 </div>
                 """)
         
@@ -718,4 +709,13 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         show_error=True
-    )
+    )
+
+# Updated requirements.txt should include:
+# torch>=2.0.0
+# transformers>=4.30.0
+# gradio
+# numpy
+# accelerate
+# spaces
+# peft  # For LoRA adapter support