A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 09:45 UTC
[Bugfix] Reduce moe_sum test size to avoid OOM (#18484)
Signed-off-by: Bill Nell <bnell@redhat.com>
bnellnm committed
May 21, 2025 at 09:46 UTC
c6c10ca920f094f402f78635a234ffdcb108d78e
1 file changed
+1
-1
tests/kernels/moe/test_moe.py
+1
-1
@@ -577,7 +577,7 @@ def test_moe_align_block_size_opcheck():
577
num_tokens_post_pad))
578
579
580
-@pytest.mark.parametrize("m", [1, 33, 222, 1024 * 128])
580
+@pytest.mark.parametrize("m", [1, 33, 64, 222])
581
@pytest.mark.parametrize("topk", TOP_KS)
582
@pytest.mark.parametrize("k", [128, 511, 1024])
583
@pytest.mark.parametrize("dtype",