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

Revert "[MoE] Decouple Mega MoE from DeepEP backend" (#25317)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-05-14 16:00:41 -07:00
committed by GitHub
parent 78408d2300
commit 67096f48bf
8 changed files with 33 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/dsv4/test_deepseek_v4_flash_fp4_megamoe_b200.py
+8 -2
View File
@@ -28,12 +28,18 @@ 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",
@@ -75,7 +81,7 @@ class TestDSV4FlashFP4B200W4A8MegaMoE(ServerSanityMixin, CustomTestCase):
"4",
"--enable-dp-attention",
"--moe-a2a-backend",
"megamoe",
"deepep",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
@@ -116,7 +122,7 @@ class TestDSV4FlashFP4B200W4A4MegaMoE(ServerSanityMixin, CustomTestCase):
"4",
"--enable-dp-attention",
"--moe-a2a-backend",
"megamoe",
"deepep",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",