File size: 6,816 Bytes
a402b9b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 | import os
import unittest
from types import SimpleNamespace
import requests
from sglang.srt.environ import envs
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_DRAFT_MODEL_STANDALONE,
DEFAULT_TARGET_MODEL_STANDALONE,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
popen_launch_server,
)
# Standalone speculative decoding tests (FA3, Triton, FlashInfer backends)
register_cuda_ci(est_time=308, suite="stage-b-test-large-1-gpu")
GSM_DATASET_PATH = None
# Default server arguments shared across all tests
DEFAULT_SERVER_ARGS = [
"--trust-remote-code",
"--cuda-graph-max-bs",
"8",
"--speculative-algorithm",
"STANDALONE",
"--speculative-draft-model-path",
DEFAULT_DRAFT_MODEL_STANDALONE,
"--speculative-num-steps",
"4",
"--speculative-eagle-topk",
"2",
"--speculative-num-draft-tokens",
"7",
"--mem-fraction-static",
0.7,
]
# Default server arguments for V2 tests
DEFAULT_SERVER_ARGS_V2 = [
"--trust-remote-code",
"--cuda-graph-max-bs",
"8",
"--speculative-algorithm",
"STANDALONE",
"--speculative-draft-model-path",
DEFAULT_DRAFT_MODEL_STANDALONE,
"--speculative-num-steps",
"4",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"5",
"--mem-fraction-static",
0.7,
]
class TestStandaloneSpeculativeDecodingBase(CustomTestCase):
model = DEFAULT_TARGET_MODEL_STANDALONE
draft_model = DEFAULT_DRAFT_MODEL_STANDALONE
base_url = DEFAULT_URL_FOR_TEST
accuracy_threshold = 0.7 # derived tests need to override this
spec_decode_threshold = 3.6 # derived spec decoding tests need to override this
@classmethod
def get_server_args(cls):
"""Return the arguments for the server launch. Override in subclasses."""
return DEFAULT_SERVER_ARGS + ["--attention-backend", "fa3"]
@classmethod
def setUpClass(cls):
# disable deep gemm precompile to make launch server faster
# please don't do this if you want to make your inference workload faster
envs.SGLANG_JIT_DEEPGEMM_PRECOMPILE.set(False)
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
model = cls.model
cls.process = popen_launch_server(
model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=cls.get_server_args(),
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=4,
num_questions=100,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
data_path=GSM_DATASET_PATH,
)
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
# Use the appropriate metric key based on the test class
metric_key = "accuracy"
self.assertGreater(metrics[metric_key], self.accuracy_threshold)
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=}")
self.assertGreater(avg_spec_accept_length, self.spec_decode_threshold)
class TestStandaloneV2SpeculativeDecodingBase(CustomTestCase):
model = DEFAULT_TARGET_MODEL_STANDALONE
draft_model = DEFAULT_DRAFT_MODEL_STANDALONE
base_url = DEFAULT_URL_FOR_TEST
accuracy_threshold = 0.7 # derived tests need to override this
spec_decode_threshold = 3.6 # derived spec decoding tests need to override this
@classmethod
def get_server_args(cls):
"""Return the arguments for the server launch. Override in subclasses."""
return DEFAULT_SERVER_ARGS_V2 + ["--attention-backend", "fa3"]
@classmethod
def setUpClass(cls):
# disable deep gemm precompile to make launch server faster
# please don't do this if you want to make your inference workload faster
envs.SGLANG_JIT_DEEPGEMM_PRECOMPILE.set(False)
envs.SGLANG_ENABLE_JIT_DEEPGEMM.set(False)
envs.SGLANG_ENABLE_SPEC_V2.set(True) # Enable Speculative Decoding V2
model = cls.model
cls.process = popen_launch_server(
model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=cls.get_server_args(),
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
if "SGLANG_ENABLE_SPEC_V2" in os.environ:
envs.SGLANG_ENABLE_SPEC_V2.set(False)
def test_gsm8k(self):
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=4,
num_questions=100,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
data_path=GSM_DATASET_PATH,
)
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
# Use the appropriate metric key based on the test class
metric_key = "accuracy"
self.assertGreater(metrics[metric_key], self.accuracy_threshold)
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=}")
self.assertGreater(avg_spec_accept_length, self.spec_decode_threshold)
class TestStandaloneSpeculativeDecodingTriton(TestStandaloneSpeculativeDecodingBase):
@classmethod
def get_server_args(cls):
return DEFAULT_SERVER_ARGS + ["--attention-backend", "triton"]
class TestStandaloneSpeculativeDecodingFlashinfer(
TestStandaloneSpeculativeDecodingBase
):
@classmethod
def get_server_args(cls):
return DEFAULT_SERVER_ARGS + ["--attention-backend", "flashinfer"]
class TestStandaloneV2SpeculativeDecodingTriton(
TestStandaloneV2SpeculativeDecodingBase
):
@classmethod
def get_server_args(cls):
return DEFAULT_SERVER_ARGS_V2 + ["--attention-backend", "triton"]
class TestStandaloneV2SpeculativeDecodingFlashinfer(
TestStandaloneV2SpeculativeDecodingBase
):
@classmethod
def get_server_args(cls):
return DEFAULT_SERVER_ARGS_V2 + ["--attention-backend", "flashinfer"]
if __name__ == "__main__":
unittest.main()
|