Upload app.py with huggingface_hub

app.py

@@ -15,9 +15,10 @@ model = AutoModelForCausalLM.from_pretrained(
         load_in_4bit=True,
         bnb_4bit_compute_dtype=torch.bfloat16,
         bnb_4bit_quant_type="nf4",
+        bnb_4bit_use_double_quant=True,
     ),
     device_map="auto",
-    torch_dtype=torch.bfloat16,
+    dtype=torch.bfloat16,
 )
 print("Model loaded!")