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

[CI] Improve github summary & enable fa3 for more models (#5796)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-04-27 15:29:46 -07:00
committed by GitHub
parent ded04b2e0a
commit daed453e84
6 changed files with 26 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
test/srt/test_bench_one_batch.py
View File

@@ -4,7 +4,6 @@ from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_MOE_MODEL_NAME_FOR_TEST,
CustomTestCase,
get_bool_env_var,
is_in_ci,
run_bench_one_batch,
write_github_step_summary,
@@ -12,15 +11,15 @@ from sglang.test.test_utils import (
class TestBenchOneBatch(CustomTestCase):
def test_bs1(self):
def test_bs1_default(self):
output_throughput = run_bench_one_batch(
DEFAULT_MODEL_NAME_FOR_TEST, ["--cuda-graph-max-bs", "2"]
)
if is_in_ci():
write_github_step_summary(
f"### test_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
f"### test_bs1_default (llama-3.1-8b)\n"
f"output_throughput: {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 135)
@@ -32,9 +31,9 @@ class TestBenchOneBatch(CustomTestCase):
if is_in_ci():
write_github_step_summary(
f"### test_moe_tp2_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
f"output_throughput: {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 124)
self.assertGreater(output_throughput, 125)
def test_torch_compile_tp2_bs1(self):
output_throughput = run_bench_one_batch(
@@ -45,9 +44,9 @@ class TestBenchOneBatch(CustomTestCase):
if is_in_ci():
write_github_step_summary(
f"### test_torch_compile_tp2_bs1\n"
f"output_throughput : {output_throughput:.2f} token/s\n"
f"output_throughput: {output_throughput:.2f} token/s\n"
)
self.assertGreater(output_throughput, 225)
self.assertGreater(output_throughput, 220)
if __name__ == "__main__":