chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[DeepSeek-V32]Update nightly performance benchmark (#15308)

Browse Source
This commit is contained in:
Baizhou Zhang
2025-12-17 11:25:31 -08:00
committed by GitHub
parent 53e151945a
commit 8451e22758
1 changed files with 14 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
test/nightly/test_deepseek_v32_perf.py
View File

@@ -20,7 +20,7 @@ class TestNightlyDeepseekV32Performance(unittest.TestCase):
# Define variant configurations
cls.variants = [
{
"name": "basic",
"name": "dp",
"other_args": [
"--trust-remote-code",
"--tp",
@@ -33,7 +33,7 @@ class TestNightlyDeepseekV32Performance(unittest.TestCase):
],
},
{
"name": "mtp",
"name": "dp+mtp",
"other_args": [
"--trust-remote-code",
"--tp",
@@ -56,36 +56,31 @@ class TestNightlyDeepseekV32Performance(unittest.TestCase):
],
},
{
"name": "nsa",
"name": "tp",
"other_args": [
"--trust-remote-code",
"--tp",
"8",
"--dp",
"8",
"--enable-dp-attention",
"--attention-backend",
"nsa",
"--nsa-prefill-backend",
"flashmla_sparse",
"--nsa-decode-backend",
"flashmla_kv",
"--model-loader-extra-config",
'{"enable_multithread_load": true}',
],
},
{
"name": "pure_tp",
"name": "tp+mtp",
"other_args": [
"--trust-remote-code",
"--tp",
"8",
"--attention-backend",
"nsa",
"--nsa-prefill-backend",
"flashmla_sparse",
"--nsa-decode-backend",
"flashmla_kv",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"3",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"4",
"--mem-frac",
"0.7",
"--model-loader-extra-config",
'{"enable_multithread_load": true}',
],