update app

app.py

@@ -106,7 +106,7 @@ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
 
 model = AutoModel.from_pretrained(
     model_name,
-    _attn_implementation="flash_attention_2", # prebuilt wheels @ (strangertoolshf/flash_attention_2_wheelhouse)
+    _attn_implementation="kernels-community/flash-attn2",
     trust_remote_code=True,
     use_safetensors=True,
 ).to(device).eval()