EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Add accuracy and latency tests of eagle into CI (#3027)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-01-21 02:55:14 -08:00
committed by GitHub
parent ec1c21cdc4
commit a4331cd260
7 changed files with 186 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
test/srt/test_bench_one_batch.py
View File

@@ -5,24 +5,46 @@ from sglang.test.test_utils import (
DEFAULT_MOE_MODEL_NAME_FOR_TEST,
is_in_ci,
run_bench_one_batch,
write_github_step_summary,
)
class TestBenchOneBatch(unittest.TestCase):
def test_default(self):
def test_bs1(self):
output_throughput = run_bench_one_batch(DEFAULT_MODEL_NAME_FOR_TEST, [])
if is_in_ci():
write_github_step_summary(
f"### test_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 135)
def test_moe_default(self):
def test_moe_tp2_bs1(self):
output_throughput = run_bench_one_batch(
DEFAULT_MOE_MODEL_NAME_FOR_TEST, ["--tp", "2"]
)
if is_in_ci():
write_github_step_summary(
f"### test_moe_tp2_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 125)
def test_torch_compile_tp2_bs1(self):
output_throughput = run_bench_one_batch(
DEFAULT_MODEL_NAME_FOR_TEST,
["--tp", "2", "--enable-torch-compile", "--cuda-graph-max-bs", "2"],
)
if is_in_ci():
write_github_step_summary(
f"### test_torch_compile_tp2_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 240)
if __name__ == "__main__":
unittest.main()