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

Update CI threshold & Improve code style (#2159)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-11-24 06:29:38 -08:00
committed by GitHub
parent e3938b2f9c
commit 5652c56535
8 changed files with 126 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
test/srt/test_bench_serving.py
View File

@@ -20,7 +20,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 2850)
self.assertGreater(res["output_throughput"], 3350)
def test_offline_throughput_non_stream_small_batch_size(self):
res = run_bench_serving(
@@ -47,7 +47,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 2900)
self.assertGreater(res["output_throughput"], 3350)
def test_offline_throughput_without_chunked_prefill(self):
res = run_bench_serving(
@@ -74,7 +74,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 2950)
self.assertGreater(res["output_throughput"], 3450)
def test_offline_throughput_default_fp8(self):
res = run_bench_serving(
@@ -85,7 +85,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 3200)
self.assertGreater(res["output_throughput"], 3850)
def test_online_latency_default(self):
res = run_bench_serving(
@@ -109,7 +109,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 1900)
self.assertGreater(res["output_throughput"], 2150)
def test_moe_offline_throughput_without_radix_cache(self):
res = run_bench_serving(
@@ -120,7 +120,7 @@ class TestBenchServing(unittest.TestCase):
)
if is_in_ci():
self.assertGreater(res["output_throughput"], 1950)
self.assertGreater(res["output_throughput"], 2150)
if __name__ == "__main__":