Skip to content

[NVIDIA] Disable the split-k from cutlass MLA #17745

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 2 commits into from
Closed

[NVIDIA] Disable the split-k from cutlass MLA #17745

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c752af6
Work around the split-k issue from cutlass kernel
kaixih May 6, 2025
74ead0d
Format
kaixih May 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion csrc/attention/mla/cutlass_mla_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,7 +119,7 @@ typename T::Fmha::Arguments args_from_options(
{static_cast<ElementOut*>(out.data_ptr()), stride_O,
static_cast<ElementAcc*>(nullptr), stride_LSE},
hw_info,
-1, // split_kv
1, // split_kv
nullptr, // is_var_split_kv
};
// TODO(kaixih@nvidia): When split_kv=-1 and is_var_split_kv=false, we compute
Expand Down
4 changes: 3 additions & 1 deletion tests/kernels/test_cutlass_mla_decode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,9 @@ def test_cutlass_mla_decode(dtype: torch.dtype, mean_seq_len: int, bs: int,
pack_factor = 128 // block_size
block_num = ((block_num + pack_factor - 1) // pack_factor) * pack_factor

q = torch.randn(bs, h_q, d)
# Amplify input values to ensure test coverage of edge cases where CUTLASS
# kernel errors occur with split_k settings.
q = torch.randn(bs, h_q, d) * 100
block_table = torch.randint(0,
bs * block_num, (bs, block_num),
dtype=torch.int32)
Expand Down