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

[CI] Trim redundant nightly test registrations (#34070)

Browse Source 
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
This commit is contained in:
Liangsheng Yin
2026-08-08 01:42:46 -07:00
committed by GitHub
co-authored by Baizhou Zhang
parent dd5d82bead
commit f6a6f5bf1e
24 changed files with 2 additions and 3571 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/backends/test_deepseek_r1_fp8_trtllm_backend.py
-90
View File
@@ -1,90 +0,0 @@
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_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
try_cached_model,
)
register_cuda_ci(est_time=3600, suite="nightly-8-gpu-b200", nightly=True)
FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
SERVER_LAUNCH_TIMEOUT = 1000
class TestDeepseekR1Fp8Flashinfer(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = try_cached_model(FULL_DEEPSEEK_V3_MODEL_PATH)
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--trust-remote-code",
"--disable-radix-cache",
"--max-running-requests",
"512",
"--chunked-prefill-size",
"8192",
"--mem-fraction-static",
"0.9",
"--cuda-graph-max-bs-decode",
"128",
"--max-prefill-tokens",
"8192",
"--kv-cache-dtype",
"fp8_e4m3",
"--quantization",
"fp8",
"--tensor-parallel-size",
"8",
"--data-parallel-size",
"1",
"--expert-parallel-size",
"1",
"--scheduler-recv-interval",
"10",
"--stream-interval",
"10",
"--attention-backend",
"trtllm_mla",
"--fp8-gemm-backend",
"flashinfer_trtllm",
"--moe-runner-backend",
"flashinfer_trtllm",
"--enable-symm-mem",
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
]
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
other_args=other_args,
)
@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=512,
num_threads=512,
)
metrics = run_eval(args)
print(f"Eval accuracy of GSM8K: {metrics=}")
self.assertGreater(metrics["score"], 0.92)
if __name__ == "__main__":
unittest.main()
test/registered/backends/test_qwen3_fp4_trtllm_gen_moe.py
-69
View File
@@ -1,69 +0,0 @@
import unittest
from types import SimpleNamespace
from sglang.srt.utils import get_device_sm, 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,
)
# modelopt_fp4 requires SM 100+ (Blackwell)
register_cuda_ci(est_time=300, suite="nightly-1-gpu", nightly=True)
@unittest.skipIf(
get_device_sm() < 100, "Test requires CUDA SM 100 or higher (Blackwell)"
)
class TestFlashinferTrtllmGenMoeBackend(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "nvidia/Qwen3-30B-A3B-NVFP4"
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=[
"--moe-runner-backend",
"flashinfer_trtllm",
"--quantization",
"modelopt_fp4",
"--trust-remote-code",
"--disable-radix-cache",
"--max-running-requests",
"1024",
"--chunked-prefill-size",
"16384",
"--mem-fraction-static",
"0.89",
"--max-prefill-tokens",
"16384",
],
)
@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=1319,
num_threads=1319,
num_shots=8,
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["score"], 0.88)
if __name__ == "__main__":
unittest.main()