updated with Flashattention

app.py

@@ -16,6 +16,7 @@ token = os.environ["HF_TOKEN"]
 
 
 model = AutoModelForCausalLM.from_pretrained(
+    use_cache=False,attn_implementation="flash_attention_2",
     "microsoft/Phi-3-mini-128k-instruct", token=token,trust_remote_code=True
 )
 tok = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct", token=token)