29 lines
833 B
Python
29 lines
833 B
Python
import unittest
|
|
|
|
from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
|
|
from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
|
|
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
|
|
|
|
MODEL = "openai/gpt-oss-20b"
|
|
|
|
register_cuda_ci(est_time=198, stage="base-b", runner_config="1-gpu-large")
|
|
register_cpu_ci(est_time=1602, suite="stage-b-test-cpu-intel")
|
|
|
|
|
|
class TestSWARadixCacheKL(KLDivergenceMixin, DefaultServerBase):
|
|
model = MODEL
|
|
kl_div_thres = 0.02 # it was 0.002
|
|
kl_div_decode_max_new_tokens = 2048
|
|
server_env = {"SGLANG_ENABLE_RANK_CONSENSUS_CHECKER": "1"}
|
|
other_args = [
|
|
"--tp-size",
|
|
"1",
|
|
"--mem-fraction-static",
|
|
"0.70",
|
|
"--cuda-graph-backend-prefill=disabled",
|
|
]
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|