Update app.py

app.py

@@ -167,64 +167,34 @@ def extract_coordinates(text):
     return [(0.5, 0.5)]
 
 def cpu_inference(conversation, model, tokenizer, processor):
-    """
-    Inference function untuk CPU with better error handling
-    """
     try:
-        # Apply chat template
-        text = processor.apply_chat_template(
-            conversation, 
-            tokenize=False, 
-            add_generation_prompt=True
-        )
-        
-        # Get image from conversation
-        image = conversation[1]["content"][0]["image"]
-        
-        # Process inputs with proper padding
         inputs = processor(
-            text=[text], 
-            images=[image], 
+            conversation,
             return_tensors="pt",
-            padding=True,  # Enable padding
-            truncation=True,  # Enable truncation for long texts
-            max_length=512  # Set reasonable max length
+            padding=True,
+            truncation=True
         )
-        
-        # Generate response with proper error handling
+
         with torch.no_grad():
-            try:
-                outputs = model.generate(
-                    **inputs,
-                    max_new_tokens=256,
-                    do_sample=True,
-                    temperature=0.3,
-                    top_p=0.8,
-                    pad_token_id=tokenizer.eos_token_id if tokenizer.eos_token_id else tokenizer.pad_token_id
-                )
-            except Exception as e:
-                logger.error(f"Generation error: {e}")
-                # Try with simpler parameters
-                outputs = model.generate(
-                    **inputs,
-                    max_new_tokens=128,
-                    do_sample=False,
-                    pad_token_id=tokenizer.eos_token_id if tokenizer.eos_token_id else 0
-                )
-        
-        # Decode response
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=256,
+                do_sample=True,
+                temperature=0.3,
+                top_p=0.8,
+                pad_token_id=tokenizer.eos_token_id or tokenizer.pad_token_id or 0
+            )
+
         generated_ids = outputs[0][inputs["input_ids"].shape[1]:]
         response = tokenizer.decode(generated_ids, skip_special_tokens=True)
-        
-        # Extract coordinates
         coordinates = extract_coordinates(response)
-        
+
         return {
             "topk_points": coordinates,
             "response": response,
             "success": True
         }
-        
+
     except Exception as e:
         logger.error(f"Inference error: {e}")
         return {