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

ci: migrate 2-GPU tests to test/registered/ (#16529)

Browse Source
This commit is contained in:
Alison Shao
2026-01-07 20:28:16 -08:00
committed by GitHub
parent ab7d5829cd
commit 63cc97f4ef
15 changed files with 84 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/srt/models/test_kimi_linear_models.py
-46
View File
@@ -1,46 +0,0 @@
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
class TestKimiLinear(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "moonshotai/Kimi-Linear-48B-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,
other_args=["--tp-size", "2", "--trust-remote"],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.88)
if __name__ == "__main__":
unittest.main()
test/srt/models/test_nvidia_nemotron_nano_v2.py
-96
View File
@@ -1,96 +0,0 @@
import unittest
from sglang.srt.utils import is_blackwell
from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, DefaultServerBase):
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
gsm8k_accuracy_thres = 0.87
other_args = ["--max-mamba-cache-size", "256"]
class TestNvidiaNemotronNanoV2BF16PP(GSM8KMixin, DefaultServerBase):
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
gsm8k_accuracy_thres = 0.87
other_args = ["--max-mamba-cache-size", "256", "--pp-size", "2"]
class TestNvidiaNemotronNanoV2FP8(GSM8KMixin, DefaultServerBase):
gsm8k_accuracy_thres = 0.87
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8"
other_args = ["--max-mamba-cache-size", "256"]
@unittest.skipIf(not is_blackwell(), "NVFP4 only supported on blackwell")
class TestNvidiaNemotronNanoV2NVFP4(GSM8KMixin, DefaultServerBase):
gsm8k_accuracy_thres = 0.855
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2-NVFP4"
other_args = ["--max-mamba-cache-size", "256"]
@unittest.skip(
"STANDALONE speculative decoding does not yet support target and draft models "
"with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
)
class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, DefaultServerBase):
gsm8k_accuracy_thres = 0.87
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
other_args = [
"--speculative-algorithm",
"STANDALONE",
"--speculative-num-steps",
"2",
"--speculative-eagle-topk",
"3",
"--speculative-num-draft-tokens",
"5",
"--speculative-draft-model-path",
"meta-llama/Llama-3.2-1B",
"--speculative-draft-load-format",
"dummy",
"--max-running-requests",
"8",
"--max-total-tokens",
"2048",
"--json-model-override-args",
'{"vocab_size": 131072}',
]
@unittest.skip(
"STANDALONE speculative decoding does not yet support target and draft models "
"with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
)
class TestNvidiaNemotronNanoV2SpeculativeDecodingBF16Cache(
GSM8KMixin, DefaultServerBase
):
gsm8k_accuracy_thres = 0.87
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
other_args = [
"--speculative-algorithm",
"STANDALONE",
"--speculative-num-steps",
"2",
"--speculative-eagle-topk",
"3",
"--speculative-num-draft-tokens",
"5",
"--speculative-draft-model-path",
"meta-llama/Llama-3.2-1B",
"--speculative-draft-load-format",
"dummy",
"--max-running-requests",
"8",
"--max-total-tokens",
"2048",
"--json-model-override-args",
'{"vocab_size": 131072}',
"--mamba-ssm-dtype",
"bfloat16",
]
if __name__ == "__main__":
unittest.main()