chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD] MORI-EP inter kernel type switch (#18437)

Browse Source 
Co-authored-by: HAI <hixiao@gmail.com>
This commit is contained in:
Duyi-Wang
2026-02-16 12:59:39 +08:00
committed by GitHub
parent bc79a64d3a
commit 5ddc84e33e
4 changed files with 18 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
python/sglang/srt/layers/moe/token_dispatcher/moriep.py
View File

@@ -85,9 +85,21 @@ class EpDispatchConfig:
rdma_block_num: int
def get_ep_dispatch_configs():
def get_ep_dispatch_configs(num_max_dispatch_tokens_per_rank: int = 4096):
import mori
# Selects the inter-node kernel. `InterNodeV1LL` is used if `num_max_dispatch_tokens_per_rank`
# is less than or equal to the threshold, otherwise `InterNodeV1` is used. The threshold defaults to 256.
inter_kernel_switch_threshold = get_int_env_var(
"SGLANG_MORI_DISPATCH_INTER_KERNEL_SWITCH_THRESHOLD", 256
)
inter_kernel_type = (
mori.ops.EpDispatchCombineKernelType.InterNodeV1LL
if num_max_dispatch_tokens_per_rank <= inter_kernel_switch_threshold
else mori.ops.EpDispatchCombineKernelType.InterNodeV1
)
return {
EpMode.INTRA_NODE: EpDispatchConfig(
kernel_type=mori.ops.EpDispatchCombineKernelType.IntraNode,
@@ -96,7 +108,7 @@ def get_ep_dispatch_configs():
rdma_block_num=0,
),
EpMode.INTER_NODE: EpDispatchConfig(
kernel_type=mori.ops.EpDispatchCombineKernelType.InterNodeV1,
kernel_type=inter_kernel_type,
warp_num_per_block=8,
block_num=64,
rdma_block_num=32,
@@ -130,7 +142,7 @@ def init_mori_op(
)
mode = EpMode.INTRA_NODE if world_size <= 8 else EpMode.INTER_NODE
cfg = get_ep_dispatch_configs()[mode]
cfg = get_ep_dispatch_configs(num_max_dispatch_tokens_per_rank)[mode]
kernel_type = cfg.kernel_type
warp_num_per_block = cfg.warp_num_per_block