test_retract_decode.py 1.56 KB
Newer Older
Lianmin Zheng's avatar
Lianmin Zheng committed
1
import os
2
3
4
import unittest
from types import SimpleNamespace

5
from sglang.srt.utils import kill_process_tree
6
7
8
9
10
11
12
13
14
15
16
17
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,
    popen_launch_server,
)


class TestRetractDecode(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
Lianmin Zheng's avatar
Lianmin Zheng committed
18
19
        os.environ["SGLANG_TEST_RETRACT"] = "1"

20
21
22
23
24
25
26
27
        cls.model = DEFAULT_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
        )

    @classmethod
    def tearDownClass(cls):
28
        kill_process_tree(cls.process.pid)
29
30
31
32
33
34
35
36
37
38
39

    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)
40
        self.assertGreaterEqual(metrics["score"], 0.65)
41
42


Lianmin Zheng's avatar
Lianmin Zheng committed
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
class TestRetractDecodeChunkCache(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
        os.environ["SGLANG_TEST_RETRACT"] = "1"

        cls.model = DEFAULT_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,
            other_args=["--disable-radix-cache", "--chunked-prefill-size", 128],
        )


58
59
if __name__ == "__main__":
    unittest.main()