Update app.py

app.py

@@ -17,7 +17,7 @@ This is a demo of [`MBZUAI-Paris/Atlas-Chat-9B`](https://huggingface.co/MBZUAI-P
 
 MAX_MAX_NEW_TOKENS = 2048
 DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "2024"))
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
@@ -27,8 +27,8 @@ model = AutoModelForCausalLM.from_pretrained(
     model_id,
     device_map="auto",
     torch_dtype=torch.bfloat16,
+    _attn_implementation="eager"
 )
-model.config.sliding_window = 4096
 model.eval()