80 lines
2.1 KiB
Python
80 lines
2.1 KiB
Python
"""
|
|
Usage:
|
|
python3 -m unittest test_intel_amx_attention_backend.TestIntelAMXAttnBackend.test_mmlu
|
|
"""
|
|
|
|
import unittest
|
|
from types import SimpleNamespace
|
|
|
|
from sglang.srt.utils import kill_process_tree
|
|
from sglang.test.run_eval import run_eval
|
|
from sglang.test.test_utils import (
|
|
DEFAULT_MLA_MODEL_NAME_FOR_TEST,
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
DEFAULT_URL_FOR_TEST,
|
|
CustomTestCase,
|
|
is_in_ci,
|
|
popen_launch_server,
|
|
run_bench_one_batch,
|
|
)
|
|
|
|
|
|
class TestIntelAMXAttnBackend(CustomTestCase):
|
|
def test_latency(self):
|
|
prefill_latency, decode_throughput, decode_latency = run_bench_one_batch(
|
|
DEFAULT_MLA_MODEL_NAME_FOR_TEST,
|
|
[
|
|
"--attention-backend",
|
|
"intel_amx",
|
|
"--mem-fraction-static",
|
|
"0.05",
|
|
"--disable-radix",
|
|
"--trust-remote-code",
|
|
"--batch-size",
|
|
"4",
|
|
],
|
|
)
|
|
|
|
print(f"{prefill_latency=}")
|
|
print(f"{decode_throughput=}")
|
|
print(f"{decode_latency=}")
|
|
|
|
if is_in_ci():
|
|
self.assertGreater(decode_throughput, 10)
|
|
|
|
def test_mmlu(self):
|
|
model = DEFAULT_MLA_MODEL_NAME_FOR_TEST
|
|
base_url = DEFAULT_URL_FOR_TEST
|
|
process = popen_launch_server(
|
|
model,
|
|
base_url,
|
|
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
other_args=[
|
|
"--attention-backend",
|
|
"intel_amx",
|
|
"--mem-fraction-static",
|
|
"0.05",
|
|
"--disable-radix",
|
|
"--trust-remote-code",
|
|
"--disable-overlap-schedule",
|
|
],
|
|
)
|
|
|
|
try:
|
|
args = SimpleNamespace(
|
|
base_url=base_url,
|
|
model=model,
|
|
eval_name="mmlu",
|
|
num_examples=64,
|
|
num_threads=32,
|
|
)
|
|
|
|
metrics = run_eval(args)
|
|
self.assertGreater(metrics["score"], 0.45)
|
|
finally:
|
|
kill_process_tree(process.pid)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|