143 lines
4.5 KiB
Python
143 lines
4.5 KiB
Python
import unittest
|
|
|
|
from nightly_utils import NightlyBenchmarkRunner
|
|
|
|
from sglang.test.test_utils import DEFAULT_URL_FOR_TEST, _parse_int_list_env
|
|
|
|
DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2-Exp"
|
|
PROFILE_DIR = "performance_profiles_deepseek_v32"
|
|
|
|
|
|
class TestNightlyDeepseekV32Basic(unittest.TestCase):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.model = DEEPSEEK_V32_MODEL_PATH
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
cls.batch_sizes = [1, 1, 8, 16, 64]
|
|
cls.input_lens = tuple(_parse_int_list_env("NIGHTLY_INPUT_LENS", "4096"))
|
|
cls.output_lens = tuple(_parse_int_list_env("NIGHTLY_OUTPUT_LENS", "512"))
|
|
cls.other_args = [
|
|
"--trust-remote-code",
|
|
"--tp",
|
|
"8",
|
|
"--dp",
|
|
"8",
|
|
"--enable-dp-attention",
|
|
]
|
|
cls.runner = NightlyBenchmarkRunner(PROFILE_DIR, cls.__name__, cls.base_url)
|
|
cls.runner.setup_profile_directory()
|
|
|
|
def test_bench_one_batch(self):
|
|
results, success = self.runner.run_benchmark_for_model(
|
|
model_path=self.model,
|
|
batch_sizes=self.batch_sizes,
|
|
input_lens=self.input_lens,
|
|
output_lens=self.output_lens,
|
|
other_args=self.other_args,
|
|
variant="basic",
|
|
)
|
|
|
|
self.runner.add_report(results)
|
|
self.runner.write_final_report()
|
|
|
|
if not success:
|
|
raise AssertionError(
|
|
f"Benchmark failed for {self.model} with basic configuration"
|
|
)
|
|
|
|
|
|
class TestNightlyDeepseekV32MTP(unittest.TestCase):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.model = DEEPSEEK_V32_MODEL_PATH
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
cls.batch_sizes = [1, 1, 8, 16, 64]
|
|
cls.input_lens = tuple(_parse_int_list_env("NIGHTLY_INPUT_LENS", "4096"))
|
|
cls.output_lens = tuple(_parse_int_list_env("NIGHTLY_OUTPUT_LENS", "512"))
|
|
cls.other_args = [
|
|
"--trust-remote-code",
|
|
"--tp",
|
|
"8",
|
|
"--dp",
|
|
"8",
|
|
"--enable-dp-attention",
|
|
"--speculative-algorithm",
|
|
"EAGLE",
|
|
"--speculative-num-steps",
|
|
"3",
|
|
"--speculative-eagle-topk",
|
|
"1",
|
|
"--speculative-num-draft-tokens",
|
|
"4",
|
|
"--mem-frac",
|
|
"0.7",
|
|
]
|
|
cls.runner = NightlyBenchmarkRunner(PROFILE_DIR, cls.__name__, cls.base_url)
|
|
cls.runner.setup_profile_directory()
|
|
|
|
def test_bench_one_batch(self):
|
|
results, success = self.runner.run_benchmark_for_model(
|
|
model_path=self.model,
|
|
batch_sizes=self.batch_sizes,
|
|
input_lens=self.input_lens,
|
|
output_lens=self.output_lens,
|
|
other_args=self.other_args,
|
|
variant="mtp",
|
|
)
|
|
|
|
self.runner.add_report(results)
|
|
self.runner.write_final_report()
|
|
|
|
if not success:
|
|
raise AssertionError(
|
|
f"Benchmark failed for {self.model} with MTP configuration"
|
|
)
|
|
|
|
|
|
class TestNightlyDeepseekV32NSA(unittest.TestCase):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.model = DEEPSEEK_V32_MODEL_PATH
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
cls.batch_sizes = [1, 1, 8, 16, 64]
|
|
cls.input_lens = tuple(_parse_int_list_env("NIGHTLY_INPUT_LENS", "4096"))
|
|
cls.output_lens = tuple(_parse_int_list_env("NIGHTLY_OUTPUT_LENS", "512"))
|
|
cls.other_args = [
|
|
"--trust-remote-code",
|
|
"--tp",
|
|
"8",
|
|
"--dp",
|
|
"8",
|
|
"--enable-dp-attention",
|
|
"--attention-backend",
|
|
"nsa",
|
|
"--nsa-prefill-backend",
|
|
"flashmla_sparse",
|
|
"--nsa-decode-backend",
|
|
"flashmla_kv",
|
|
]
|
|
cls.runner = NightlyBenchmarkRunner(PROFILE_DIR, cls.__name__, cls.base_url)
|
|
cls.runner.setup_profile_directory()
|
|
|
|
def test_bench_one_batch(self):
|
|
results, success = self.runner.run_benchmark_for_model(
|
|
model_path=self.model,
|
|
batch_sizes=self.batch_sizes,
|
|
input_lens=self.input_lens,
|
|
output_lens=self.output_lens,
|
|
other_args=self.other_args,
|
|
variant="nsa",
|
|
)
|
|
|
|
self.runner.add_report(results)
|
|
self.runner.write_final_report()
|
|
|
|
if not success:
|
|
raise AssertionError(
|
|
f"Benchmark failed for {self.model} with NSA configuration"
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|