test_deepep_small.py 10.8 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
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.test_utils import (
    DEFAULT_MODEL_NAME_FOR_TEST_MLA,
    DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
    DEFAULT_URL_FOR_TEST,
    CustomTestCase,
    popen_launch_server,
)


class TestPureDP(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-dp-attention",
                "--dp",
                "4",
                "--enable-deepep-moe",
                "--cuda-graph-max-bs",
                "128",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
38
                "512",
39
40
                "--mem-fraction-static",
                "0.5",
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

61
        self.assertGreater(metrics["accuracy"], 0.60)
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83


class TestHybridDPTP(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-dp-attention",
                "--dp",
                "2",
                "--enable-deepep-moe",
                "--cuda-graph-max-bs",
                "128",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
84
                "256",
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

105
        self.assertGreater(metrics["accuracy"], 0.60)
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


class TestTP(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-deepep-moe",
                "--cuda-graph-max-bs",
                "128",
                "--max-running-requests",
                "128",
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

146
        self.assertGreater(metrics["accuracy"], 0.60)
147
148


149
@unittest.skip("covered in test_deepep_large.py")
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
class TestNoGatherdBuffer(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-dp-attention",
                "--dp",
                "4",
                "--moe-dense-tp-size",
                "1",
                "--enable-dp-lm-head",
                "--enable-deepep-moe",
                "--cuda-graph-max-bs",
                "32",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
173
                "512",
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

194
        self.assertGreater(metrics["accuracy"], 0.60)
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219


class TestTBO(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-dp-attention",
                "--dp",
                "4",
                "--moe-dense-tp-size",
                "1",
                "--enable-deepep-moe",
                "--enable-two-batch-overlap",
                "--cuda-graph-max-bs",
                "128",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
220
                "512",
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

241
        self.assertGreater(metrics["accuracy"], 0.60)
242
243


244
@unittest.skip("covered in TestMTPWithTBO")
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
class TestMTP(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--trust-remote-code",
                "--tp",
                "4",
                "--enable-dp-attention",
                "--dp",
                "2",
                "--enable-dp-lm-head",
                "--enable-deepep-moe",
                "--speculative-algo",
                "NEXTN",
                "--speculative-draft",
                DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
                "--speculative-num-steps",
                "2",
                "--speculative-eagle-topk",
                "3",
                "--speculative-num-draft-tokens",
                "3",
                "--cuda-graph-max-bs",
                "32",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
276
                "64",
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

        self.assertGreater(metrics["accuracy"], 0.60)

        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"###test_gsm8k (deepseek-v3 mtp + dp + tbo):\n"
            f"accuracy={metrics['accuracy']=:.3f}\n"
            f"{avg_spec_accept_length=:.3f}\n"
        )
        self.assertGreater(avg_spec_accept_length, 2.1)


class TestMTPWithTBO(CustomTestCase):
    @classmethod
    def setUpClass(cls):
        import os

        cls.model = DEFAULT_MODEL_NAME_FOR_TEST_MLA
        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=[
                "--tp-size",
                "4",
                "--enable-dp-attention",
                "--dp-size",
                "4",
                "--enable-two-batch-overlap",
                "--enable-deepep-moe",
                "--trust-remote-code",
                "--speculative-algorithm",
                "NEXTN",
                "--speculative-num-steps",
                "2",
                "--speculative-eagle-topk",
                "3",
                "--speculative-num-draft-tokens",
                "3",
                "--speculative-draft",
                DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN,
                "--chunked-prefill-size",
                "256",
                "--cuda-graph-max-bs",
                "32",
                "--max-running-requests",
Cheng Wan's avatar
Cheng Wan committed
346
                "128",
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
            ],
        )

    @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_few_shot_gsm8k(args)
        print(metrics)

        self.assertGreater(metrics["accuracy"], 0.60)

        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"###test_gsm8k (deepseek-v3 mtp + dp + tbo):\n"
            f"accuracy={metrics['accuracy']=:.3f}\n"
            f"{avg_spec_accept_length=:.3f}\n"
        )
        self.assertGreater(avg_spec_accept_length, 2.1)


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