Update app.py

app.py

@@ -61,9 +61,12 @@ def load_models():
 
     base_mod = AutoModelForCausalLM.from_pretrained(
         BASE_ID,
-        device_map="auto",
-        token=HF_TOKEN
+        token=HF_TOKEN,
+        device_map="auto",           # deja que Accelerate decida
+        quantization_config=quant_config,
+        offload_folder="/tmp/offload"   # ruta temporal de escritura
     )
+)
     norm_tok = AutoTokenizer.from_pretrained(ADAPTER_ID, use_fast=True, token=HF_TOKEN)
     norm_mod = PeftModel.from_pretrained(
         base_mod,