[MoE] Decouple Mega MoE from DeepEP backend (#25406)
This commit is contained in:
@@ -100,7 +100,6 @@ class TestB200FlashCP(DSV4FlashAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "1024",
|
||||
}
|
||||
|
||||
|
||||
@@ -116,7 +116,6 @@ class TestB200ProCP(DSV4ProAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
|
||||
}
|
||||
|
||||
|
||||
@@ -102,7 +102,6 @@ class TestB300FlashCP(DSV4FlashAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "1024",
|
||||
}
|
||||
|
||||
|
||||
@@ -118,7 +118,6 @@ class TestB300ProCP(DSV4ProAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
|
||||
}
|
||||
|
||||
|
||||
@@ -24,9 +24,7 @@ DSV4_FLASH_MODEL_PATH = "sgl-project/DeepSeek-V4-Flash-FP8"
|
||||
|
||||
DSV4_FLASH_BASE_ENV = {
|
||||
"SGLANG_ENABLE_SPEC_V2": "1",
|
||||
"SGLANG_OPT_USE_TOPK_V2": "1",
|
||||
"SGLANG_DSV4_FP4_EXPERTS": "0",
|
||||
"SGLANG_JIT_DEEPGEMM_PRECOMPILE": "0",
|
||||
}
|
||||
|
||||
DSV4_FLASH_SERVER_ARGS = [
|
||||
|
||||
@@ -41,9 +41,7 @@ HONGLOUMENG_PATH = os.environ.get(
|
||||
|
||||
DSV4_PRO_BASE_ENV = {
|
||||
"SGLANG_ENABLE_SPEC_V2": "1",
|
||||
"SGLANG_OPT_USE_TOPK_V2": "1",
|
||||
"SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2": "1",
|
||||
"SGLANG_JIT_DEEPGEMM_PRECOMPILE": "0",
|
||||
}
|
||||
|
||||
DSV4_PRO_SERVER_ARGS = [
|
||||
|
||||
@@ -100,7 +100,6 @@ class TestGB300FlashCP(DSV4FlashAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "1024",
|
||||
}
|
||||
|
||||
|
||||
@@ -118,7 +118,6 @@ class TestGB300ProCP(DSV4ProAime25TestBase):
|
||||
DEEPEP_LARGE_SMS_CONFIG,
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
|
||||
}
|
||||
|
||||
|
||||
@@ -112,7 +112,6 @@ class TestH200Fp8FlashCP(DSV4FlashAime25TestBase):
|
||||
]
|
||||
EXTRA_ENV = {
|
||||
**H200_FP8_ENV,
|
||||
"SGLANG_OPT_USE_JIT_INDEXER_METADATA": "1",
|
||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "1024",
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user