b294bd4bc7
Co-authored-by: Zhang, Mingxu <mingxu.zhang@intel.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: MingxuZh <109504044+MingxuZh@users.noreply.github.com>
124 lines
3.4 KiB
Python
124 lines
3.4 KiB
Python
"""
|
|
Usage:
|
|
python3 -m unittest test_intel_amx_attention_backend.TestIntelAMXAttnBackend.test_latency_default_model
|
|
"""
|
|
|
|
import unittest
|
|
from types import SimpleNamespace
|
|
|
|
from sglang.srt.utils import kill_process_tree
|
|
from sglang.test.ci.ci_register import register_cpu_ci
|
|
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_MLA_MODEL_NAME_FOR_TEST,
|
|
DEFAULT_MODEL_NAME_FOR_TEST,
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
DEFAULT_URL_FOR_TEST,
|
|
CustomTestCase,
|
|
intel_amx_benchmark,
|
|
is_in_ci,
|
|
popen_launch_server,
|
|
)
|
|
|
|
register_cpu_ci(est_time=685, suite="base-b-tp-test-cpu")
|
|
|
|
|
|
class TestIntelAMXAttnBackend(CustomTestCase):
|
|
|
|
@intel_amx_benchmark(
|
|
extra_args=["--batch-size", "4", "--mem-fraction-static", "0.3"],
|
|
min_throughput=10,
|
|
)
|
|
def test_latency_mla_model(self):
|
|
return DEFAULT_MLA_MODEL_NAME_FOR_TEST
|
|
|
|
@intel_amx_benchmark(
|
|
extra_args=["--batch-size", "4", "--mem-fraction-static", "0.3"],
|
|
min_throughput=40,
|
|
)
|
|
def test_latency_default_model(self):
|
|
return DEFAULT_MODEL_NAME_FOR_TEST
|
|
|
|
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",
|
|
"0.3",
|
|
"--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)
|
|
if is_in_ci():
|
|
self.assertGreater(metrics["score"], 0.45)
|
|
finally:
|
|
kill_process_tree(process.pid)
|
|
|
|
|
|
class TestDPAttention(CustomTestCase):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.model = DEFAULT_MLA_MODEL_NAME_FOR_TEST
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
other_args = [
|
|
"--trust-remote-code",
|
|
"--disable-radix-cache",
|
|
"--attention-backend",
|
|
"intel_amx",
|
|
"--mem-fraction-static",
|
|
"0.3",
|
|
"--disable-overlap-schedule",
|
|
"--tp",
|
|
"2",
|
|
"--enable-dp-attention",
|
|
"--dp",
|
|
"2",
|
|
]
|
|
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_dp_attention_DP2TP2(self):
|
|
args = SimpleNamespace(
|
|
num_shots=5,
|
|
data_path=None,
|
|
num_questions=32,
|
|
parallel=32,
|
|
max_new_tokens=512,
|
|
host="http://127.0.0.1",
|
|
port=int(self.base_url.split(":")[-1]),
|
|
)
|
|
metrics = run_eval_few_shot_gsm8k(args)
|
|
print(f"Eval accuracy of GSM8K: {metrics=}")
|
|
|
|
self.assertGreater(metrics["accuracy"], 0.7)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|