Support FA3 as Attention backend by using --attention-backend fa3 (#4680)

Co-authored-by: qsong <qsong@linkedin.com> Co-authored-by: qingquansong <ustcsqq@gmail.com>
2025-03-23 23:28:11 -07:00
parent af6535e7aa
commit 5d7edc8e55
5 changed files with 622 additions and 1 deletions
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -770,7 +770,7 @@ class ServerArgs:
        parser.add_argument(
            "--attention-backend",
            type=str,
-            choices=["flashinfer", "triton", "torch_native"],
+            choices=["flashinfer", "triton", "torch_native", "fa3"],
            default=ServerArgs.attention_backend,
            help="Choose the kernels for attention layers.",
        )