Update interface.py

interface.py

@@ -22,20 +22,27 @@ tokenizer = AutoTokenizer.from_pretrained(model_path)
 model = AutoModelForCausalLM.from_pretrained(model_path)
 # No movemos el modelo al dispositivo aquí
 
+from decorators import spaces
+
 @spaces.GPU(duration=100)
-def generate_analysis(prompt, max_length=MAX_LENGTH, device=None):
+def generate_analysis(prompt, max_length=1024, device=None):
     try:
         if device is None:
             device = torch.device('cpu')
+        
+        # Mover el modelo al dispositivo adecuado (GPU o CPU)
         if next(model.parameters()).device != device:
             model.to(device)
+        
+        # Preparar los datos de entrada en el dispositivo correcto
         input_ids = tokenizer.encode(prompt, return_tensors='pt').to(device)
         max_gen_length = min(max_length + input_ids.size(1), model.config.max_position_embeddings)
 
+        # Generar el texto
         generated_ids = model.generate(
             input_ids=input_ids,
             max_length=max_gen_length,
-            temperature=TEMPERATURE,
+            temperature=0.7,
             num_return_sequences=1,
             no_repeat_ngram_size=2,
             early_stopping=True
@@ -44,6 +51,8 @@ def generate_analysis(prompt, max_length=MAX_LENGTH, device=None):
         output_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
         analysis = output_text[len(prompt):].strip()
         return analysis
+    except RuntimeError as e:
+        return f"Error durante la ejecución: {str(e)}"
     except Exception as e:
         return f"Ocurrió un error durante el análisis: {e}"