Fix the default arguments of bench_offline_throughput.py & simplify detokenizer manager (#2042)
This commit is contained in:
@@ -11,16 +11,24 @@ 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)
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=True, chunked_prefill_size=32, enable_overlap=True
|
||||
)
|
||||
|
||||
def test_no_radix_attention_no_chunked_prefill(self):
|
||||
run_mmlu_test(disable_radix_cache=True, chunked_prefill_size=-1)
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=True, chunked_prefill_size=-1, enable_overlap=True
|
||||
)
|
||||
|
||||
def test_radix_attention_chunked_prefill(self):
|
||||
run_mmlu_test(disable_radix_cache=False, chunked_prefill_size=32)
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=False, chunked_prefill_size=32, enable_overlap=True
|
||||
)
|
||||
|
||||
def test_radix_attention_no_chunked_prefill(self):
|
||||
run_mmlu_test(disable_radix_cache=False, chunked_prefill_size=-1)
|
||||
run_mmlu_test(
|
||||
disable_radix_cache=False, chunked_prefill_size=-1, enable_overlap=True
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
Reference in New Issue
Block a user