adds flash attention

app.py

@@ -12,6 +12,7 @@ try:
         "openai/gpt-oss-20b",
         torch_dtype="auto",
         device_map="auto",
+        attn_implementation="kernel-community/vllm-flash-attention3"
     )
     tokenizer = AutoTokenizer.from_pretrained("openai/gpt-oss-20b")