test_data_parallelism.py 2.11 KB
Newer Older
1
import time
2
3
4
import unittest
from types import SimpleNamespace

5
6
import requests

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


18
class TestDataParallelism(CustomTestCase):
19
20
21
22
23
24
25
26
    @classmethod
    def setUpClass(cls):
        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,
27
            other_args=["--dp", 2],
28
29
30
31
        )

    @classmethod
    def tearDownClass(cls):
32
        kill_process_tree(cls.process.pid)
33
34
35
36
37
38
39
40
41
42
43

    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)
44
        self.assertGreaterEqual(metrics["score"], 0.65)
45

46
47
    def test_update_weight(self):
        response = requests.post(
Chayenne's avatar
Chayenne committed
48
            self.base_url + "/update_weights_from_disk",
49
50
51
52
53
54
55
            json={"model_path": DEFAULT_MODEL_NAME_FOR_TEST},
        )

        # check if the response is 200
        assert response.status_code == 200

        # pause a few seconds then send again
56
        time.sleep(1)
57

58
        response = requests.post(
Chayenne's avatar
Chayenne committed
59
            self.base_url + "/update_weights_from_disk",
60
61
62
63
64
65
            json={"model_path": DEFAULT_MODEL_NAME_FOR_TEST},
        )

        # check if the response is 200
        assert response.status_code == 200

66
    def test_get_memory_pool_size(self):
67
68
        # use `get_server_info` instead since `get_memory_pool_size` is merged into `get_server_info`
        response = requests.get(self.base_url + "/get_server_info")
69
70
        assert response.status_code == 200

71
        time.sleep(1)
72

73
        response = requests.get(self.base_url + "/get_server_info")
74
75
        assert response.status_code == 200

76
77
78

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