diff --git a/test/registered/radix_cache/test_swa_radix_cache_kl.py b/test/registered/radix_cache/test_swa_radix_cache_kl.py index 795ad9d9e..97afbd45b 100644 --- a/test/registered/radix_cache/test_swa_radix_cache_kl.py +++ b/test/registered/radix_cache/test_swa_radix_cache_kl.py @@ -11,7 +11,7 @@ register_cuda_ci(est_time=151, stage="base-b", runner_config="1-gpu-large") class TestSWARadixCacheKL(KLDivergenceMixin, DefaultServerBase): model = MODEL - kl_div_thres = 0.002 + kl_div_thres = 0.02 # it was 0.002 kl_div_decode_max_new_tokens = 2048 other_args = [ "--tp-size",