Update app.py

app.py

@@ -68,7 +68,8 @@ model = AutoModelForCausalLM.from_pretrained(
     model_id,
     device_map="auto",
     torch_dtype=torch.bfloat16,
-    trust_remote_code=True
+    trust_remote_code=True,
+    attn_implementation="eager",
 )
 
 @spaces.GPU