EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Cutlass MLA: Disable split kv due to https://github.com/NVIDIA/cutlass/issues/2274 (#6101)

Browse Source
This commit is contained in:
Trevor Morris
2025-05-08 18:44:30 -07:00
committed by GitHub
parent cec98f1034
commit 0ab3f437ab
2 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
sgl-kernel/tests/test_cutlass_mla.py
View File

@@ -67,7 +67,7 @@ def test_cutlass_mla_decode(
pack_factor = 128 // block_size
block_num = ((block_num + pack_factor - 1) // pack_factor) * pack_factor
q = torch.randn(bs, h_q, d)
q = torch.randn(bs, h_q, d) * 100.0
block_table = torch.randint(0, bs * block_num, (bs, block_num), dtype=torch.int32)
kv_cache = torch.randn(block_table.numel(), block_size, d)