Integrate triton moe kernel (#7689)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-07-07 11:05:49 +08:00
parent ea3e7ffec7
commit 253454de9b
7 changed files with 697 additions and 54 deletions
--- a/python/sglang/srt/managers/schedule_batch.py
+++ b/python/sglang/srt/managers/schedule_batch.py
@@ -101,6 +101,7 @@ GLOBAL_SERVER_ARGS_KEYS = [
    "triton_attention_reduce_in_fp32",
    "num_reserved_decode_tokens",
    "weight_loader_disable_mmap",
+    "enable_triton_kernel_moe",
 ]

 # Put some global args for easy access