Files
sglang/test/srt/test_flash_attention_4.py
Johnny 252dc4e112 [NVIDIA] FA3/FA4 Fix (#11606)
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
2025-10-19 17:10:10 -07:00

57 lines
1.6 KiB
Python

import unittest
from types import SimpleNamespace
from sglang.srt.environ import envs
from sglang.srt.utils import get_device_sm, kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
popen_launch_server,
)
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
class TestFlashAttention4(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--trust-remote-code",
"--mem-fraction-static",
"0.8",
"--prefill-attention-backend",
"fa4",
]
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_gsm8k(self):
args = SimpleNamespace(
num_shots=4,
data_path=None,
num_questions=100,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
print(metrics)
self.assertGreater(metrics["accuracy"], 0.65)
if __name__ == "__main__":
unittest.main()