test_deepseek_v3_fp4_4gpu.py 6.34 KB
Newer Older
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
64
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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
import unittest
from types import SimpleNamespace

import requests

from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.send_one import BenchArgs, send_one_prompt
from sglang.test.test_utils import (
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
    DEFAULT_URL_FOR_TEST,
    CustomTestCase,
    is_in_ci,
    popen_launch_server,
    write_github_step_summary,
)

FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"


class TestDeepseekV3FP4(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
        cls.base_url = DEFAULT_URL_FOR_TEST
        other_args = [
            "--tp",
            "4",
            "--attention-backend",
            "trtllm_mla",
            "--moe-runner-backend",
            "flashinfer_trtllm",
            "--quantization",
            "modelopt_fp4",
        ]
        cls.process = popen_launch_server(
            cls.model,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=other_args,
        )

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

    def test_a_gsm8k(
        self,
    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
        args = SimpleNamespace(
            num_shots=8,
            data_path=None,
            num_questions=1319,
            parallel=1319,
            max_new_tokens=512,
            host="http://127.0.0.1",
            port=int(self.base_url.split(":")[-1]),
        )
        metrics = run_eval_few_shot_gsm8k(args)
        print(f"{metrics=}")

        if is_in_ci():
            write_github_step_summary(
                f"### test_gsm8k (deepseek-v3-fp4)\n" f'{metrics["accuracy"]=:.3f}\n'
            )
            self.assertGreater(metrics["accuracy"], 0.935)

    def test_bs_1_speed(self):
        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
        acc_length, speed = send_one_prompt(args)

        print(f"{speed=:.2f}")

        if is_in_ci():
            write_github_step_summary(
                f"### test_bs_1_speed (deepseek-v3-fp4)\n" f"{speed=:.2f} token/s\n"
            )
            self.assertGreater(speed, 75)


class TestDeepseekV3FP4MTP(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
        cls.base_url = DEFAULT_URL_FOR_TEST
        other_args = [
            "--tp",
            "4",
            "--attention-backend",
            "trtllm_mla",
            "--moe-runner-backend",
            "flashinfer_trtllm",
            "--quantization",
            "modelopt_fp4",
            "--speculative-algorithm",
            "EAGLE",
            "--speculative-num-steps",
            "3",
            "--speculative-eagle-topk",
            "1",
            "--speculative-num-draft-tokens",
            "4",
        ]
        cls.process = popen_launch_server(
            cls.model,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=other_args,
        )

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

    def test_a_gsm8k(
        self,
    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
        requests.get(self.base_url + "/flush_cache")

        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_few_shot_gsm8k(args)
        print(f"{metrics=}")

        server_info = requests.get(self.base_url + "/get_server_info")
        avg_spec_accept_length = server_info.json()["internal_states"][0][
            "avg_spec_accept_length"
        ]
        print(f"{avg_spec_accept_length=}")

        if is_in_ci():
            write_github_step_summary(
                f"### test_gsm8k (deepseek-v3-fp4 mtp)\n"
                f'{metrics["accuracy"]=:.3f}\n'
                f"{avg_spec_accept_length=:.2f}\n"
            )
            self.assertGreater(metrics["accuracy"], 0.935)
            self.assertGreater(avg_spec_accept_length, 2.9)

    def test_bs_1_speed(self):
        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
        acc_length, speed = send_one_prompt(args)

        print(f"{acc_length=:.2f} {speed=:.2f}")

        if is_in_ci():
            write_github_step_summary(
                f"### test_bs_1_speed (deepseek-v3-fp4 mtp)\n"
                f"{acc_length=:.2f}\n"
                f"{speed=:.2f} token/s\n"
            )
            self.assertGreater(acc_length, 2.9)
            self.assertGreater(speed, 130)


163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
        cls.base_url = DEFAULT_URL_FOR_TEST
        other_args = [
            "--tp",
            "4",
            "--ep",
            "4",
            "--attention-backend",
            "trtllm_mla",
            "--moe-runner-backend",
            "flashinfer_cutlass",
            "--quantization",
            "modelopt_fp4",
        ]
        cls.process = popen_launch_server(
            cls.model,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=other_args,
        )

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

    def test_a_gsm8k(
        self,
    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
        args = SimpleNamespace(
            num_shots=8,
            data_path=None,
            num_questions=1319,
            parallel=1319,
            max_new_tokens=512,
            host="http://127.0.0.1",
            port=int(self.base_url.split(":")[-1]),
        )
        metrics = run_eval_few_shot_gsm8k(args)
        print(f"{metrics=}")

        if is_in_ci():
            write_github_step_summary(
                f"### test_gsm8k (deepseek-v3-fp4-cutlass-moe)\n"
                f'{metrics["accuracy"]=:.3f}\n'
            )
            self.assertGreater(metrics["accuracy"], 0.935)


214
215
if __name__ == "__main__":
    unittest.main()