EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[ci]use H20 to run disaggregation test (#11543)

Browse Source
This commit is contained in:
Hank Han
2025-10-17 02:42:42 +08:00
committed by GitHub
parent 0975ba99bc
commit 0dd6cf16ba
6 changed files with 75 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/srt/run_suite.py
View File

@@ -163,9 +163,7 @@ suites = {
TestFile("lora/test_lora_llama4.py", 400),
TestFile("test_deepseek_v3_basic.py", 275),
TestFile("test_deepseek_v3_mtp.py", 275),
TestFile("test_disaggregation_different_tp.py", 600),
TestFile("test_disaggregation_hybrid_attention.py", 200),
TestFile("test_disaggregation_pp.py", 140),
],
"per-commit-4-gpu-b200": [
# TestFile("test_gpt_oss_4gpu.py", 600),
@@ -182,6 +180,8 @@ suites = {
TestFile("test_deepseek_v32_basic.py", 275),
],
"per-commit-8-gpu-h20": [
TestFile("test_disaggregation_different_tp.py", 600),
TestFile("test_disaggregation_pp.py", 140),
TestFile("quant/test_w4a8_deepseek_v3.py", 371),
],
"vllm_dependency_test": [

9
test/srt/test_disaggregation_different_tp.py
View File

@@ -9,6 +9,7 @@ from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST_MLA,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
popen_launch_pd_server,
try_cached_model,
)
@@ -19,7 +20,7 @@ class TestDisaggregationMooncakePrefillLargerTP(TestDisaggregationBase):
# Temporarily disable JIT DeepGEMM
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST_MLA)
# Non blocking start servers
cls.start_prefill()
@@ -90,7 +91,7 @@ class TestDisaggregationMooncakeDecodeLargerTP(TestDisaggregationBase):
# Temporarily disable JIT DeepGEMM
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST_MLA)
# Non blocking start servers
cls.start_prefill()
@@ -161,7 +162,7 @@ class TestDisaggregationMooncakeMHAPrefillLargerTP(TestDisaggregationBase):
# Temporarily disable JIT DeepGEMM
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
# Non blocking start servers
cls.start_prefill()
@@ -232,7 +233,7 @@ class TestDisaggregationMooncakeMHADecodeLargerTP(TestDisaggregationBase):
# Temporarily disable JIT DeepGEMM
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
# Non blocking start servers
cls.start_prefill()

3
test/srt/test_disaggregation_pp.py
View File

@@ -8,6 +8,7 @@ from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
popen_launch_pd_server,
try_cached_model,
)
@@ -15,7 +16,7 @@ class TestDisaggregationPPAccuracy(TestDisaggregationBase):
@classmethod
def setUpClass(cls):
super().setUpClass()
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
# Non blocking start servers
cls.start_prefill()