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

[Refactor] Refactor DeepEP dispatcher (#22822)

Browse Source 
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
This commit is contained in:
Артем Савкин
2026-05-18 04:36:42 +03:00
committed by GitHub
co-authored by gemini-code-assist[bot] Cheng Wan
parent 5147de26e4
commit a080358cac
30 changed files with 302 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/manual/layers/moe/test_moe_runners_4gpu.py
+2 -1
View File
@@ -55,7 +55,6 @@ class TestMoERunner4GPU(CustomTestCase):
},
"moe_runner_cutlass_w4a8_deepep_ll": {
"model": "tencent/DeepSeek-V3.1-Terminus-W4AFP8", # FP8 W8A8 MoE model
"env_overrides": {"SGLANG_DEEPEP_BF16_DISPATCH": "1"},
"other_args": [
"--trust-remote-code",
"--moe-runner-backend",
@@ -64,6 +63,8 @@ class TestMoERunner4GPU(CustomTestCase):
"deepep",
"--deepep-mode",
"low_latency",
"--deepep-dispatcher-output-dtype",
"bf16",
"--attention-backend",
"triton",
"--sampling-backend",
test/manual/test_w4a8_deepseek_v3.py
+2 -1
View File
@@ -192,6 +192,8 @@ class TestDeepseekV3W4Afp8DeepepAutoMtp(CustomTestCase):
"deepep",
"--deepep-mode",
"auto",
"--deepep-dispatcher-output-dtype",
"bf16",
"--dp",
"8",
"--enable-dp-attention",
@@ -215,7 +217,6 @@ class TestDeepseekV3W4Afp8DeepepAutoMtp(CustomTestCase):
other_args=other_args,
env={
**os.environ,
"SGLANG_DEEPEP_BF16_DISPATCH": "1",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
},
)
test/registered/4-gpu-models/test_deepseek_v3_cutedsl_4gpu.py
+8 -6
View File
@@ -46,12 +46,14 @@ class TestDeepseekR1Nvfp4CuteDSLDeepEP(CustomTestCase):
"modelopt_fp4",
"--attention-backend",
"trtllm_mla",
"--moe-a2a-backend",
"deepep",
"--moe-runner-backend",
"flashinfer_cutedsl",
"--moe-a2a-backend",
"deepep",
"--deepep-mode",
"low_latency",
"--deepep-dispatcher-output-dtype",
"bf16",
]
cls.process = popen_launch_server(
cls.model,
@@ -60,7 +62,6 @@ class TestDeepseekR1Nvfp4CuteDSLDeepEP(CustomTestCase):
other_args=other_args,
env={
**os.environ,
"SGLANG_DEEPEP_BF16_DISPATCH": "1",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
"SGLANG_MOE_NVFP4_DISPATCH": "0",
},
@@ -117,12 +118,14 @@ class TestDummyWithSBO(CustomTestCase):
"modelopt_fp4",
"--attention-backend",
"trtllm_mla",
"--moe-a2a-backend",
"deepep",
"--moe-runner-backend",
"flashinfer_cutedsl",
"--moe-a2a-backend",
"deepep",
"--deepep-mode",
"low_latency",
"--deepep-dispatcher-output-dtype",
"bf16",
"--json-model-override-args",
'{"num_hidden_layers": 1, "first_k_dense_replace": 0, "n_routed_experts": 24}',
"--enable-single-batch-overlap",
@@ -136,7 +139,6 @@ class TestDummyWithSBO(CustomTestCase):
other_args=other_args,
env={
**os.environ,
"SGLANG_DEEPEP_BF16_DISPATCH": "1",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "256",
"SGLANG_MOE_NVFP4_DISPATCH": "0",
},