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

Use spec v2 by default (#21062)

Browse Source
This commit is contained in:
Qiaolin Yu
2026-04-29 13:40:42 -07:00
committed by GitHub
parent c3ab5bec7d
commit 79dbfe4505
27 changed files with 413 additions and 207 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/ep/test_deepep_large.py
+44 -42
View File
@@ -3,6 +3,7 @@ from types import SimpleNamespace
import requests
from sglang.srt.environ import envs
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.run_eval import run_eval
@@ -86,48 +87,49 @@ class TestDeepseekMTP(CustomTestCase):
def setUpClass(cls):
cls.model = DEFAULT_DEEPEP_MODEL_NAME_FOR_TEST
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--trust-remote-code",
"--tp",
"8",
"--enable-dp-attention",
"--dp",
"8",
"--moe-dense-tp-size",
"1",
"--enable-dp-lm-head",
"--moe-a2a-backend",
"deepep",
"--moe-runner-backend",
"deep_gemm",
"--enable-two-batch-overlap",
"--ep-num-redundant-experts",
"32",
"--ep-dispatch-algorithm",
"dynamic",
"--eplb-algorithm",
"deepseek",
"--cuda-graph-bs",
"64", # TODO: increase it to 128 when TBO is supported in draft_extend
"--max-running-requests",
"512",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"1",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"2",
"--disable-radix-cache",
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
],
)
with envs.SGLANG_ENABLE_SPEC_V2.override(False):
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--trust-remote-code",
"--tp",
"8",
"--enable-dp-attention",
"--dp",
"8",
"--moe-dense-tp-size",
"1",
"--enable-dp-lm-head",
"--moe-a2a-backend",
"deepep",
"--moe-runner-backend",
"deep_gemm",
"--enable-two-batch-overlap",
"--ep-num-redundant-experts",
"32",
"--ep-dispatch-algorithm",
"dynamic",
"--eplb-algorithm",
"deepseek",
"--cuda-graph-bs",
"64", # TODO: increase it to 128 when TBO is supported in draft_extend
"--max-running-requests",
"512",
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"1",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"2",
"--disable-radix-cache",
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
],
)
@classmethod
def tearDownClass(cls):