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

[MoE] Decouple Mega MoE from DeepEP backend (#24884)

Browse Source
This commit is contained in:
Yuhao Yang
2026-05-15 02:01:44 +08:00
committed by GitHub
parent 3fc60e5748
commit 37f030a0de
8 changed files with 44 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/dsv4/test_deepseek_v4_flash_fp4_megamoe_b200.py
+2 -8
View File
@@ -28,18 +28,12 @@ SERVER_LAUNCH_TIMEOUT = 3600
_W4A8_MEGAMOE_ENV = {
"SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE": "1",
"SGLANG_OPT_FIX_MEGA_MOE_MEMORY": "1",
"SGLANG_OPT_FIX_NEXTN_MEGA_MOE": "1",
"SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK": "4096",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "0",
}
_W4A4_MEGAMOE_ENV = {
"SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE": "1",
"SGLANG_OPT_FIX_MEGA_MOE_MEMORY": "1",
"SGLANG_OPT_FIX_NEXTN_MEGA_MOE": "1",
"SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK": "4096",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "0",
"SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_FP4_ACTS": "1",
@@ -81,7 +75,7 @@ class TestDSV4FlashFP4B200W4A8MegaMoE(ServerSanityMixin, CustomTestCase):
"4",
"--enable-dp-attention",
"--moe-a2a-backend",
"deepep",
"megamoe",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
@@ -122,7 +116,7 @@ class TestDSV4FlashFP4B200W4A4MegaMoE(ServerSanityMixin, CustomTestCase):
"4",
"--enable-dp-attention",
"--moe-a2a-backend",
"deepep",
"megamoe",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",