75 lines
2.0 KiB
Python
75 lines
2.0 KiB
Python
"""
|
|
Usage:
|
|
python3 -m unittest test_ascend_attention_backend.TestAscendAttnBackend.test_gsm8k
|
|
"""
|
|
|
|
import unittest
|
|
from types import SimpleNamespace
|
|
from urllib.parse import urlparse
|
|
|
|
from sglang.srt.utils import kill_process_tree
|
|
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
|
|
from sglang.test.run_eval import run_eval
|
|
from sglang.test.test_utils import (
|
|
DEFAULT_MODEL_NAME_FOR_TEST,
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
DEFAULT_URL_FOR_TEST,
|
|
CustomTestCase,
|
|
is_in_ci,
|
|
popen_launch_server,
|
|
run_bench_offline_throughput,
|
|
)
|
|
|
|
|
|
class TestAscendAttnBackend(CustomTestCase):
|
|
def test_latency(self):
|
|
output_throughput = run_bench_offline_throughput(
|
|
DEFAULT_MODEL_NAME_FOR_TEST,
|
|
[
|
|
"--attention-backend",
|
|
"ascend",
|
|
],
|
|
)
|
|
|
|
print(f"{output_throughput=}")
|
|
|
|
if is_in_ci():
|
|
self.assertGreater(output_throughput, 18)
|
|
|
|
def test_gsm8k(self):
|
|
model = DEFAULT_MODEL_NAME_FOR_TEST
|
|
base_url = DEFAULT_URL_FOR_TEST
|
|
url = urlparse(base_url)
|
|
process = popen_launch_server(
|
|
model,
|
|
base_url,
|
|
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
other_args=[
|
|
"--attention-backend",
|
|
"ascend",
|
|
"--mem-fraction-static",
|
|
0.8,
|
|
],
|
|
)
|
|
|
|
try:
|
|
args = SimpleNamespace(
|
|
num_shots=5,
|
|
data_path=None,
|
|
num_questions=1319,
|
|
max_new_tokens=512,
|
|
parallel=128,
|
|
host=f"http://{url.hostname}",
|
|
port=int(url.port),
|
|
)
|
|
|
|
metrics = run_eval_few_shot_gsm8k(args)
|
|
self.assertGreaterEqual(metrics["accuracy"], 0.62)
|
|
self.assertLessEqual(metrics["latency"], 150)
|
|
finally:
|
|
kill_process_tree(process.pid)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|