Prueba copiando falcon3

src/model_load.py

@@ -13,10 +13,7 @@ def load_model():
     # Cargar tokenizer y modelo de Hugging Face
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     model = AutoModelForCausalLM.from_pretrained(model_name,
-                                                 device_map="auto",
-                                                 quantization_config=BitsAndBytesConfig(load_in_8bit=True),
-                                                 torch_dtype="auto",
-                                                 max_memory=max_memory)
+                                                 torch_dtype=torch.bfloat16).to("cuda")
     
     # Crear pipeline de generación de texto
     text_generation_pipeline = pipeline(