test_mla.py 1.32 KB
Newer Older
Ke Bao's avatar
Ke Bao committed
1
2
3
import unittest
from types import SimpleNamespace

4
from sglang.srt.utils import kill_process_tree
Ke Bao's avatar
Ke Bao committed
5
6
7
8
9
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,
10
    CustomTestCase,
Ke Bao's avatar
Ke Bao committed
11
12
13
14
    popen_launch_server,
)


15
class TestMLA(CustomTestCase):
Ke Bao's avatar
Ke Bao committed
16
17
18
19
20
21
22
23
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MLA_MODEL_NAME_FOR_TEST
        cls.base_url = DEFAULT_URL_FOR_TEST
        cls.process = popen_launch_server(
            cls.model,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
24
25
26
            other_args=[
                "--trust-remote-code",
                "--enable-torch-compile",
27
28
                "--torch-compile-max-bs",
                "4",
29
30
                "--chunked-prefill-size",
                "256",
31
            ],
Ke Bao's avatar
Ke Bao committed
32
33
34
35
        )

    @classmethod
    def tearDownClass(cls):
36
        kill_process_tree(cls.process.pid)
Ke Bao's avatar
Ke Bao committed
37

Ke Bao's avatar
Ke Bao committed
38
39
40
41
42
43
44
45
46
47
    def test_mgsm_en(self):
        args = SimpleNamespace(
            base_url=self.base_url,
            model=self.model,
            eval_name="mgsm_en",
            num_examples=None,
            num_threads=1024,
        )

        metrics = run_eval(args)
48
        self.assertGreater(metrics["score"], 0.8)
Ke Bao's avatar
Ke Bao committed
49

Ke Bao's avatar
Ke Bao committed
50
51
52

if __name__ == "__main__":
    unittest.main()