Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
test_pp_single_node.py 4 KB
Newer Older
Ying Sheng's avatar
[PP] Add pipeline parallelism (#5724)  
Ying Sheng committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63 
"""
Usage:
python3 -m unittest test_pp_single_node.TestPPAccuracy.test_gsm8k
python3 -m unittest test_pp_single_node.TestFixedBugs.test_chunked_prefill_with_small_bs
"""

import os
import time
import unittest
from types import SimpleNamespace

import requests

from sglang.bench_one_batch_server import BenchArgs as OneBatchBenchArgs
from sglang.srt.server_args import ServerArgs
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.runners import DEFAULT_PROMPTS
from sglang.test.test_utils import (
    DEFAULT_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
    DEFAULT_URL_FOR_TEST,
    popen_launch_server,
    run_bench_one_batch_server,
)


class TestPPAccuracy(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
        # These config helps find a leak.
        os.environ["SGLANG_IS_IN_CI"] = "1"
        cls.base_url = "http://127.0.0.1:23333"
        cls.process = popen_launch_server(
            DEFAULT_MODEL_NAME_FOR_TEST,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=[
                "--pp-size",
                4,
                "--disable-overlap-schedule",
                "--chunked-prefill-size",
                256,
            ],
        )

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

    def test_gsm8k(self):
        args = SimpleNamespace(
            num_shots=5,
            data_path=None,
            num_questions=200,
            max_new_tokens=512,
            parallel=128,
            host="http://127.0.0.1",
            port=int(self.base_url.split(":")[-1]),
        )
        metrics = run_eval(args)
        print(f"{metrics=}")
Cheng Wan's avatar
Deferring 8 GPU test (#6102)  
Cheng Wan committed
64 
        self.assertGreater(metrics["accuracy"], 0.74)
Ying Sheng's avatar
[PP] Add pipeline parallelism (#5724)  
Ying Sheng committed
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143 
        # Wait a little bit so that the memory check happens.
        time.sleep(5)


# class TestPPAccuracyFlashInfer(unittest.TestCase):
#     @classmethod
#     def setUpClass(cls):
#         # These config helps find a leak.
#         os.environ["SGLANG_IS_IN_CI"] = "1"
#         cls.base_url = "http://127.0.0.1:23333"
#         cls.process = popen_launch_server(
#             DEFAULT_MODEL_NAME_FOR_TEST,
#             cls.base_url,
#             timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
#             other_args=[
#                 "--pp-size",
#                 4,
#                 "--disable-overlap-schedule",
#                 "--attention-backend",
#                 "flashinfer",
#                 "--chunked-prefill-size",
#                 256,
#             ],
#         )
#
#     @classmethod
#     def tearDownClass(cls):
#         kill_process_tree(cls.process.pid)
#
#     def test_gsm8k(self):
#         args = SimpleNamespace(
#             num_shots=5,
#             data_path=None,
#             num_questions=200,
#             max_new_tokens=512,
#             parallel=128,
#             host="http://127.0.0.1",
#             port=int(self.base_url.split(":")[-1]),
#         )
#         metrics = run_eval(args)
#         print(f"{metrics=}")
#
#         self.assertGreater(metrics["accuracy"], 0.75)
#         # Wait a little bit so that the memory check happens.
#         time.sleep(5)


class TestFixedBugs(unittest.TestCase):
    def test_chunked_prefill_with_small_bs(self):
        model = DEFAULT_MODEL_NAME_FOR_TEST
        server_args = ServerArgs(model_path=model)
        bench_args = OneBatchBenchArgs(
            batch_size=(1,),
            input_len=(1,),
            output_len=(1,),
            base_url=DEFAULT_URL_FOR_TEST,
        )
        other_server_args = [
            "--tp-size",
            2,
            "--pp-size",
            2,
            "--disable-overlap-schedule",
            "--chunked-prefill",
            256,
            "--max-running-requests",
            2,
        ]
        run_bench_one_batch_server(
            model,
            DEFAULT_URL_FOR_TEST,
            server_args,
            bench_args,
            other_server_args,
        )


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