Update app.py

app.py

@@ -9,7 +9,7 @@ import numpy as np
 import requests
 import torch
 import os
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 from langchain.memory import ConversationBufferMemory
 
 # Configuración del modelo de lenguaje
@@ -23,13 +23,16 @@ if not HF_TOKEN:
 
 print("🔄 Cargando modelo de lenguaje...")
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)
+bnb_config = BitsAndBytesConfig(load_in_8bit=True)  # Cargar en 8-bit para reducir memoria
+
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
-    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
     device_map="auto",
+    quantization_config=bnb_config,  # Cargar modelo en 8-bit
     token=HF_TOKEN
 ).to(device)
 
+
 # Memoria conversacional
 memory = ConversationBufferMemory()