enable auto-round quantization model (#6226)

Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
This commit is contained in:
Weiwei
2025-09-08 13:05:35 +08:00
committed by GitHub
parent b67c277f86
commit c8295d2353
8 changed files with 528 additions and 0 deletions

View File

@@ -0,0 +1,62 @@
"""
Usage:
python3 -m unittest test_autoround.TestAutoRound.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_AUTOROUND_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
class TestAutoRound(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.base_url = DEFAULT_URL_FOR_TEST
@classmethod
def tearDownClass(cls):
pass
def test_mmlu(self):
device = "auto"
for model in DEFAULT_AUTOROUND_MODEL_NAME_FOR_TEST:
with self.subTest(model=model):
print(f"\n[INFO] Launching server for model: {model}")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=["--trust-remote-code", "--quantization", "auto-round"],
device=device,
)
try:
args = SimpleNamespace(
base_url=self.base_url,
model=model,
eval_name="mmlu",
num_examples=64,
num_threads=32,
device=device,
)
metrics = run_eval(args)
if "Llama" in model:
self.assertGreaterEqual(metrics["score"], 0.6)
else:
self.assertGreaterEqual(metrics["score"], 0.26)
finally:
kill_process_tree(process.pid)
print(f"[INFO] Server for {model} stopped.")
if __name__ == "__main__":
unittest.main()