test_retract_decode.py 2.28 KB
Newer Older
1
import time
2
3
4
import unittest
from types import SimpleNamespace

5
from sglang.srt.environ import envs
6
from sglang.srt.utils import kill_process_tree
7
8
9
10
11
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,
12
    CustomTestCase,
13
14
15
16
    popen_launch_server,
)


17
class TestRetractDecode(CustomTestCase):
18
19
20
21
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST
        cls.base_url = DEFAULT_URL_FOR_TEST
22
23
24
25
        with envs.SGLANG_TEST_RETRACT.override(True):
            cls.process = popen_launch_server(
                cls.model, cls.base_url, timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
            )
26
27
28

    @classmethod
    def tearDownClass(cls):
29
        kill_process_tree(cls.process.pid)
30
31
32
33
34
35
36
37
38
39
40

    def test_mmlu(self):
        args = SimpleNamespace(
            base_url=self.base_url,
            model=self.model,
            eval_name="mmlu",
            num_examples=64,
            num_threads=32,
        )

        metrics = run_eval(args)
41
        self.assertGreaterEqual(metrics["score"], 0.65)
42
43
44
        time.sleep(1)  # wait for mem check

        assert self.process.poll() is None, "Server crashed during test"
45
46


47
class TestRetractDecodeChunkCache(CustomTestCase):
Lianmin Zheng's avatar
Lianmin Zheng committed
48
49
50
51
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST
        cls.base_url = DEFAULT_URL_FOR_TEST
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
        with envs.SGLANG_TEST_RETRACT.override(True):
            cls.process = popen_launch_server(
                cls.model,
                cls.base_url,
                timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
                other_args=["--disable-radix-cache", "--chunked-prefill-size", 128],
            )

    def test_mmlu(self):
        args = SimpleNamespace(
            base_url=self.base_url,
            model=self.model,
            eval_name="mmlu",
            num_examples=64,
            num_threads=32,
Lianmin Zheng's avatar
Lianmin Zheng committed
67
68
        )

69
70
71
72
73
74
75
76
77
78
        metrics = run_eval(args)
        self.assertGreaterEqual(metrics["score"], 0.65)
        time.sleep(1)  # wait for mem check

        assert self.process.poll() is None, "Server crashed during test"

    @classmethod
    def tearDownClass(cls):
        kill_process_tree(cls.process.pid)

Lianmin Zheng's avatar
Lianmin Zheng committed
79

80
81
if __name__ == "__main__":
    unittest.main()