support using fa4 on deepseek on blackwell (#9928)

2025-09-17 07:16:06 +08:00
parent 311de47bb7
commit a2f7218a2e
7 changed files with 136 additions and 0 deletions
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -96,6 +96,7 @@ ATTENTION_BACKEND_CHOICES = [
    # NVIDIA specific
    "cutlass_mla",
    "fa3",
+    "fa4",
    "flashinfer",
    "flashmla",
    "trtllm_mla",