minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Test] Consolidate eval accuracy test mixins into eval_accuracy_kit (#21047)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-03-26 14:26:46 -07:00
committed by GitHub
parent e5dd411f64
commit fb90c9d298
25 changed files with 276 additions and 377 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/eval/test_eval_accuracy_large.py
+7 -61
View File
@@ -4,27 +4,28 @@ python -m unittest test_eval_accuracy_large.TestEvalAccuracyLarge.test_mmlu
"""
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.run_eval import run_eval
from sglang.test.kits.eval_accuracy_kit import HumanEvalMixin, MGSMEnMixin, MMLUMixin
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_amd_ci,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
register_cuda_ci(est_time=300, suite="stage-b-test-1-gpu-small")
register_amd_ci(est_time=420, suite="stage-b-test-1-gpu-small-amd")
class TestEvalAccuracyLarge(CustomTestCase):
class TestEvalAccuracyLarge(CustomTestCase, MMLUMixin, HumanEvalMixin, MGSMEnMixin):
mmlu_score_threshold = 0.70
humaneval_score_threshold = 0.64
humaneval_score_threshold_amd = 0.60
mgsm_en_score_threshold = 0.835
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
@@ -40,61 +41,6 @@ class TestEvalAccuracyLarge(CustomTestCase):
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_mmlu(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=5000,
num_threads=1024,
)
metrics = run_eval(args)
if is_in_ci():
write_github_step_summary(f"### test_mmlu\n" f'{metrics["score"]=:.4f}\n')
self.assertGreater(metrics["score"], 0.70)
def test_human_eval(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="humaneval",
num_examples=None,
num_threads=1024,
)
metrics = run_eval(args)
if is_in_ci():
write_github_step_summary(
f"### test_human_eval\n" f'{metrics["score"]=:.4f}\n'
)
if is_in_amd_ci():
self.assertGreater(metrics["score"], 0.60)
else:
self.assertGreater(metrics["score"], 0.64)
def test_mgsm_en(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="mgsm_en",
num_examples=None,
num_threads=1024,
)
metrics = run_eval(args)
if is_in_ci():
write_github_step_summary(
f"### test_mgsm_en\n" f'{metrics["score"]=:.4f}\n'
)
self.assertGreater(metrics["score"], 0.835)
if __name__ == "__main__":
unittest.main()
test/registered/eval/test_moe_eval_accuracy_large.py
+6 -54
View File
@@ -5,27 +5,28 @@ python -m unittest test_moe_eval_accuracy_large.TestMoEEvalAccuracyLarge.test_mm
import os
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.run_eval import run_eval
from sglang.test.kits.eval_accuracy_kit import HumanEvalMixin, MGSMEnMixin, MMLUMixin
from sglang.test.test_utils import (
DEFAULT_MOE_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_amd_ci,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
register_cuda_ci(est_time=500, suite="stage-b-test-2-gpu-large")
register_amd_ci(est_time=500, suite="stage-b-test-2-gpu-large-amd")
class TestMoEEvalAccuracyLarge(CustomTestCase):
class TestMoEEvalAccuracyLarge(CustomTestCase, MMLUMixin, HumanEvalMixin, MGSMEnMixin):
mmlu_score_threshold = 0.62
humaneval_score_threshold = 0.40
mgsm_en_score_threshold = 0.61
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MOE_MODEL_NAME_FOR_TEST
@@ -56,55 +57,6 @@ class TestMoEEvalAccuracyLarge(CustomTestCase):
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_mmlu(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=5000,
num_threads=1024,
)
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.62)
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,
model=self.model,
eval_name="humaneval",
num_examples=None,
num_threads=1024,
)
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.40)
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,
model=self.model,
eval_name="mgsm_en",
num_examples=None,
num_threads=1024,
)
metrics = run_eval(args)
self.assertGreater(metrics["score"], 0.61)
if is_in_ci():
write_github_step_summary(
f"### test_mgsm_en\n" f'{metrics["score"]=:.4f}\n'
)
if __name__ == "__main__":
unittest.main()