[CI] Fix ci tests (#5769)

This commit is contained in:
Lianmin Zheng
2025-04-27 07:18:10 -07:00
committed by GitHub
parent 35ca04d2fa
commit 621e96bf9b
18 changed files with 126 additions and 295 deletions

View File

@@ -4,8 +4,8 @@ from types import SimpleNamespace
from sglang.srt.utils import is_hip, kill_process_tree
from sglang.test.run_eval import run_eval
from sglang.test.test_utils import (
DEFAULT_FP8_MODEL_NAME_FOR_ACCURACY_TEST,
DEFAULT_FP8_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST,
DEFAULT_MODEL_NAME_FOR_ACCURACY_TEST_FP8,
DEFAULT_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST_FP8,
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
class TestEvalFP8Accuracy(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_FP8_MODEL_NAME_FOR_ACCURACY_TEST
cls.model = DEFAULT_MODEL_NAME_FOR_ACCURACY_TEST_FP8
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model, cls.base_url, timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
@@ -76,7 +76,7 @@ class TestEvalFP8DynamicQuantAccuracy(CustomTestCase):
def test_mmlu_offline_only(self):
"""Test with offline quantization only."""
self._run_test(
model=DEFAULT_FP8_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST,
model=DEFAULT_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST_FP8,
other_args=[],
expected_score=0.64,
)
@@ -84,7 +84,7 @@ class TestEvalFP8DynamicQuantAccuracy(CustomTestCase):
def test_mmlu_offline_and_online_override(self):
"""Test with both offline and online quantization."""
self._run_test(
model=DEFAULT_FP8_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST,
model=DEFAULT_MODEL_NAME_FOR_DYNAMIC_QUANT_ACCURACY_TEST_FP8,
other_args=["--quantization", "w8a8_fp8"],
# inference will use sgl kernel w/ online quant override
# we observed that the accuracy is higher then offline only