chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[CI] Add Nemotron 3 Super 120B CI tests for BF16 and NVFP4 (#20575)

Browse Source
This commit is contained in:
Mohammad Miadh Angkad
2026-03-15 03:30:27 +08:00
committed by GitHub
parent 39008955ff
commit 3e643967e6
2 changed files with 212 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

106
test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py Normal file
View File

@@ -0,0 +1,106 @@
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="stage-c-test-4-gpu-b200")
NEMOTRON_3_SUPER_NVFP4_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
NEMOTRON_3_SUPER_NVFP4_ARGS = [
"--tp-size",
"4",
"--trust-remote-code",
"--reasoning-parser",
"nemotron_3",
"--tool-call-parser",
"qwen3_coder",
"--disable-radix-cache",
"--model-loader-extra-config",
'{"enable_multithread_load": true, "num_threads": 17}',
]
MTP_ARGS = [
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"3",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"4",
]
def _run_gsm8k(test_case):
args = SimpleNamespace(
model=test_case.model,
eval_name="gsm8k",
num_shots=5,
num_examples=200,
max_tokens=16000,
num_threads=200,
repeat=1,
temperature=1.0,
top_p=0.95,
base_url=test_case.base_url,
host="http://127.0.0.1",
port=int(test_case.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
test_case.assertGreaterEqual(metrics["score"], 0.96)
class TestNvidiaNemotron3SuperNVFP4(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = NEMOTRON_3_SUPER_NVFP4_MODEL
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=NEMOTRON_3_SUPER_NVFP4_ARGS,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
_run_gsm8k(self)
class TestNvidiaNemotron3SuperNVFP4MTP(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = NEMOTRON_3_SUPER_NVFP4_MODEL
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=NEMOTRON_3_SUPER_NVFP4_ARGS
+ MTP_ARGS
+ ["--max-running-requests", "200"],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
_run_gsm8k(self)
if __name__ == "__main__":
unittest.main()

106
test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py Normal file
View File

@@ -0,0 +1,106 @@
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="stage-c-test-8-gpu-h200")
NEMOTRON_3_SUPER_BF16_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
NEMOTRON_3_SUPER_BF16_ARGS = [
"--tp-size",
"8",
"--trust-remote-code",
"--reasoning-parser",
"nemotron_3",
"--tool-call-parser",
"qwen3_coder",
"--disable-radix-cache",
"--model-loader-extra-config",
'{"enable_multithread_load": true, "num_threads": 50}',
]
MTP_ARGS = [
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"3",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"4",
]
def _run_gsm8k(test_case):
args = SimpleNamespace(
model=test_case.model,
eval_name="gsm8k",
num_shots=5,
num_examples=200,
max_tokens=16000,
num_threads=200,
repeat=1,
temperature=1.0,
top_p=0.95,
base_url=test_case.base_url,
host="http://127.0.0.1",
port=int(test_case.base_url.split(":")[-1]),
)
metrics = run_eval(args)
print(f"{metrics=}")
test_case.assertGreaterEqual(metrics["score"], 0.96)
class TestNvidiaNemotron3SuperBF16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = NEMOTRON_3_SUPER_BF16_MODEL
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=NEMOTRON_3_SUPER_BF16_ARGS,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
_run_gsm8k(self)
class TestNvidiaNemotron3SuperBF16MTP(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = NEMOTRON_3_SUPER_BF16_MODEL
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=NEMOTRON_3_SUPER_BF16_ARGS
+ MTP_ARGS
+ ["--max-running-requests", "200"],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
_run_gsm8k(self)
if __name__ == "__main__":
unittest.main()