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

[CI] Trim redundant B200 test registrations (#33586)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-08-04 13:22:00 -07:00
committed by GitHub
parent aa06433709
commit 0753663b8e
9 changed files with 4 additions and 270 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/attention/unittests/dense/test_fa3.py
-1
View File
@@ -35,7 +35,6 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
run_dense_split_op_extend_case,
)
register_cuda_ci(est_time=25, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=25, stage="base-b", runner_config="1-gpu-large")
test/registered/backends/test_flashinfer_trtllm_gen_attn_backend.py
-65
View File
@@ -1,65 +0,0 @@
import os
import unittest
from types import SimpleNamespace
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
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
register_cuda_ci(est_time=300, suite="nightly-4-gpu-b200", nightly=True)
class TestFlashinferTrtllmGenAttnBackend(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "Qwen/Qwen3-Next-80B-A3B-Instruct"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
env={**os.environ, "SGLANG_ENABLE_JIT_DEEPGEMM": "False"},
other_args=[
"--attention-backend",
"trtllm_mha",
"--cuda-graph-max-bs-decode",
"512",
"--tp-size",
"4",
"--ep-size",
"4",
"--mem-fraction-static",
"0.7",
"--mamba-ssm-dtype",
"bfloat16",
"--disable-radix-cache",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="gsm8k",
api="completion",
max_tokens=512,
num_examples=200,
num_threads=128,
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["score"], 0.93)
if __name__ == "__main__":
unittest.main()
test/registered/lora/test_lora_qwen3_5_35b_a3b_logprob_diff.py
+1 -1
View File
@@ -37,7 +37,7 @@ from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=110,
stage="base-c",
runner_config="4-gpu-b200",
runner_config="4-gpu-h100",
)
BASE_MODEL = "Qwen/Qwen3.5-35B-A3B"
test/registered/lora/test_lora_qwen3_vl_30b_a3b_instruct_logprob_diff.py
+1 -1
View File
@@ -37,7 +37,7 @@ from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=110,
stage="base-c",
runner_config="4-gpu-b200",
runner_config="4-gpu-h100",
)
BASE_MODEL = "Qwen/Qwen3-VL-30B-A3B-Instruct"
test/registered/models_e2e/test_gpt_oss_4gpu_bf16.py
-1
View File
@@ -4,7 +4,6 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.gpt_oss_common import BaseTestGptOss
register_cuda_ci(est_time=220, stage="base-c", runner_config="4-gpu-h100")
register_cuda_ci(est_time=220, stage="base-c", runner_config="4-gpu-b200")
class TestGptOss4GpuBf16(BaseTestGptOss):
test/registered/models_e2e/test_qwen35_fp4_flashinfer.py
-83
View File
@@ -1,83 +0,0 @@
import unittest
import torch
from sglang.test.accuracy_test_runner import AccuracyTestParams
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.run_combined_tests import run_combined_tests
from sglang.test.test_utils import (
CustomTestCase,
ModelLaunchSettings,
)
register_cuda_ci(est_time=720, stage="base-c", runner_config="4-gpu-b200")
QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
_cuda_major = int(torch.version.cuda.split(".")[0]) if torch.version.cuda else 0
_is_sm100_cuda13 = (
torch.cuda.is_available()
and torch.cuda.get_device_capability()[0] >= 10
and _cuda_major >= 13
)
@unittest.skipUnless(_is_sm100_cuda13, "requires SM100+ GPU and CUDA 13+")
class TestQwen35FP4FlashInfer(CustomTestCase):
def test_gsm8k(self):
base_args = [
"--tp-size",
"4",
"--chunked-prefill-size",
"2048",
"--mamba-scheduler-strategy",
"extra_buffer",
"--mamba-track-interval",
"128",
"--mamba-ssm-dtype",
"bfloat16",
"--max-running-requests",
"128",
"--reasoning-parser",
"qwen3",
"--attention-backend",
"trtllm_mha",
"--quantization",
"modelopt_fp4",
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
"--linear-attn-decode-backend",
"flashinfer",
"--linear-attn-prefill-backend",
"flashinfer",
]
variants = [
ModelLaunchSettings(
QWEN35_FP4_MODEL,
extra_args=base_args,
variant="FlashInfer",
),
]
run_combined_tests(
models=variants,
test_name="Qwen3.5-397B-A17B-NVFP4",
accuracy_params=AccuracyTestParams(
dataset="gsm8k",
baseline_accuracy=ACC_THRESHOLDS[QWEN35_FP4_MODEL]["gsm8k"],
num_examples=200,
num_threads=128,
max_tokens=16000,
thinking_mode="qwen3",
temperature=0.6,
top_p=0.95,
top_k=20,
),
)
if __name__ == "__main__":
unittest.main()
test/registered/models_e2e/test_qwen35_fp4_mtp.py
+1 -29
View File
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=740, stage="base-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=400, stage="base-c", runner_config="4-gpu-b200")
QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
@@ -106,33 +106,5 @@ class TestQwen35FP4MTP(ReasoningTokenUsageMixin, CustomTestCase):
_run_mtp_gsm8k(self)
class TestQwen35FP4MTPFlashInfer(ReasoningTokenUsageMixin, CustomTestCase):
reasoning_parser_name = "qwen3"
@classmethod
def setUpClass(cls):
cls.model = QWEN35_FP4_MODEL
cls.base_url = DEFAULT_URL_FOR_TEST
cls.init_reasoning_token_verifier()
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=MTP_BASE_ARGS
+ [
"--linear-attn-decode-backend",
"flashinfer",
"--enforce-disable-flashinfer-allreduce-fusion",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
_run_mtp_gsm8k(self)
if __name__ == "__main__":
unittest.main()
test/registered/spec/eagle/test_eagle_dp_attention.py
+1 -4
View File
@@ -21,10 +21,7 @@ from sglang.test.test_utils import (
)
# EAGLE3 with DP attention (tp=2, dp=2, requires 4 GPUs).
# Per-commit EAGLE + DP-attn coverage on CUDA is provided by
# test_eagle_infer_beta_dp_attention.py (B200 4-gpu), so this H100 variant
# is gated to extra-b only.
register_cuda_ci(est_time=99, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=99, stage="base-c", runner_config="4-gpu-h100")
register_amd_ci(est_time=200, suite="stage-c-test-4-gpu-amd")
test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py
-85
View File
@@ -1,85 +0,0 @@
import unittest
from types import SimpleNamespace
import requests
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
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST_MLA,
DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
# EAGLE with DP attention on B200 (tp=2, dp=2, requires 4 B200 GPUs)
register_cuda_ci(est_time=90, stage="base-c", runner_config="4-gpu-b200")
def test_gsm8k(base_url: str, model: str):
requests.get(base_url + "/flush_cache")
args = SimpleNamespace(
base_url=base_url,
model=model,
eval_name="gsm8k",
api="completion",
max_tokens=512,
num_examples=200,
num_threads=128,
)
metrics = run_eval(args)
server_info = requests.get(base_url + "/server_info")
avg_spec_accept_length = server_info.json()["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{metrics=}")
print(f"{avg_spec_accept_length=}")
return metrics, avg_spec_accept_length
class TestEagleDPAttnServerSmall(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--tp-size",
"2",
"--dp-size",
"2",
"--enable-dp-attention",
"--speculative-draft-model-path",
DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"3",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"4",
]
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=other_args,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_a_gsm8k(self):
metrics, avg_spec_accept_length = test_gsm8k(self.base_url, self.model)
self.assertGreater(metrics["score"], 0.62)
self.assertGreater(avg_spec_accept_length, 2.7)
if __name__ == "__main__":
unittest.main()