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

Fix accuracy test (#1051)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-08-12 02:48:40 -07:00
committed by GitHub
parent 41598e0d8e
commit 0c1c72a0b4
4 changed files with 15 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
test/srt/test_eval_accuracy_large.py
View File

@@ -32,12 +32,12 @@ class TestEvalAccuracyLarge(unittest.TestCase):
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=None,
num_threads=2048,
num_examples=3000,
num_threads=1024,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.70
assert metrics["score"] >= 0.71, f"{metrics}"
def test_human_eval(self):
args = SimpleNamespace(
@@ -45,11 +45,11 @@ class TestEvalAccuracyLarge(unittest.TestCase):
model=self.model,
eval_name="humaneval",
num_examples=None,
num_threads=2048,
num_threads=1024,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.65
assert metrics["score"] >= 0.65, f"{metrics}"
def test_mgsm_en(self):
args = SimpleNamespace(
@@ -57,11 +57,11 @@ class TestEvalAccuracyLarge(unittest.TestCase):
model=self.model,
eval_name="mgsm_en",
num_examples=None,
num_threads=2048,
num_threads=1024,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.85
assert metrics["score"] >= 0.85, f"{metrics}"
if __name__ == "__main__":

8
test/srt/test_serving_throughput.py
View File

@@ -66,8 +66,8 @@ class TestServingThroughput(unittest.TestCase):
)
if os.getenv("SGLANG_IS_IN_CI", "false") == "true":
# A100 performance
assert res["output_throughput"] >= 1300
# A100 (PCIE) performance
assert res["output_throughput"] >= 1400
def test_default_without_radix_cache(self):
res = self.run_test(
@@ -77,8 +77,8 @@ class TestServingThroughput(unittest.TestCase):
)
if os.getenv("SGLANG_IS_IN_CI", "false") == "true":
# A100 performance
assert res["output_throughput"] >= 1400
# A100 (PCIE) performance
assert res["output_throughput"] >= 1450
def test_default_without_flashinfer(self):
self.run_test(