Support penalty in overlap mode; return logprob with chunked prefill; improve benchmark scripts (#3988)

Co-authored-by: SangBin Cho <rkooo567@gmail.com>
Co-authored-by: dhou-xai <dhou@x.ai>
Co-authored-by: Hanming Lu <hanming_lu@berkeley.edu>
This commit is contained in:
Lianmin Zheng
2025-03-03 00:12:04 -08:00
parent 0194948fd9
commit ac2387279e
86 changed files with 4116 additions and 2015 deletions

View File

@@ -12,7 +12,9 @@ from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
@@ -44,6 +46,9 @@ class TestEvalAccuracyLarge(unittest.TestCase):
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.71)
if is_in_ci():
write_github_step_summary(f"### test_mmlu\n" f'{metrics["score"]=:.4f}\n')
def test_human_eval(self):
args = SimpleNamespace(
base_url=self.base_url,
@@ -56,6 +61,11 @@ class TestEvalAccuracyLarge(unittest.TestCase):
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.64)
if is_in_ci():
write_github_step_summary(
f"### test_human_eval\n" f'{metrics["score"]=:.4f}\n'
)
def test_mgsm_en(self):
args = SimpleNamespace(
base_url=self.base_url,
@@ -68,6 +78,11 @@ class TestEvalAccuracyLarge(unittest.TestCase):
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.835)
if is_in_ci():
write_github_step_summary(
f"### test_mgsm_en\n" f'{metrics["score"]=:.4f}\n'
)
if __name__ == "__main__":
unittest.main()