From 341923b9820ea1dc437445e2e81644e6ba47e5b6 Mon Sep 17 00:00:00 2001 From: Aziz Date: Fri, 22 Aug 2025 19:20:59 +0200 Subject: [PATCH] fix(tests): Ensure reliable CUDA cache clearing in MoE test (#23416) Signed-off-by: AzizCode92 Signed-off-by: Michael Goin Co-authored-by: Michael Goin Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- tests/kernels/moe/test_moe.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/kernels/moe/test_moe.py b/tests/kernels/moe/test_moe.py index 1951eb0c61802..0ea9667914fd5 100644 --- a/tests/kernels/moe/test_moe.py +++ b/tests/kernels/moe/test_moe.py @@ -429,11 +429,11 @@ def test_mixtral_moe(dtype: torch.dtype, padding: bool, use_rocm_aiter: bool, vllm_moe.experts.w13_weight, (0, 128), "constant", 0)[..., 0:-128], requires_grad=False) - torch.cuda.empty_cache() vllm_moe.experts.w2_weight = Parameter(F.pad( vllm_moe.experts.w2_weight, (0, 128), "constant", 0)[..., 0:-128], requires_grad=False) + torch.cuda.synchronize() torch.cuda.empty_cache() # Run forward passes for both MoE blocks