[CI] Split test cases in CI for better load balancing (#2180)
This commit is contained in:
35
test/srt/test_no_overlap_scheduler.py
Normal file
35
test/srt/test_no_overlap_scheduler.py
Normal file
@@ -0,0 +1,35 @@
|
||||
"""
|
||||
Usage:
|
||||
python3 -m unittest test_overlap_schedule.TestOverlapSchedule.test_radix_attention_chunked_prefill
|
||||
python3 test_overlap_schedule.py
|
||||
"""
|
||||
|
||||
import unittest
|
||||
|
||||
from sglang.test.test_utils import run_mmlu_test
|
||||
|
||||
|
||||
class TestOverlapSchedule(unittest.TestCase):
|
||||
def test_no_radix_attention_chunked_prefill(self):
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=True, chunked_prefill_size=32, disable_overlap=True
|
||||
)
|
||||
|
||||
def test_no_radix_attention_no_chunked_prefill(self):
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=True, chunked_prefill_size=-1, disable_overlap=True
|
||||
)
|
||||
|
||||
def test_radix_attention_chunked_prefill(self):
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=False, chunked_prefill_size=32, disable_overlap=True
|
||||
)
|
||||
|
||||
def test_radix_attention_no_chunked_prefill(self):
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=False, chunked_prefill_size=-1, disable_overlap=True
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user