squall/flash-attention
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove contiguous checks

Browse Source
This commit is contained in:
Tri Dao 2024-08-05 14:46:46 -07:00
parent 3f1b4d38e7
commit 5d5bfbb619
1 changed files with 0 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
hopper/flash_api.cpp
View File

@ -287,10 +287,6 @@ mha_fwd(at::Tensor &q, // batch_size x seqlen_q x num_heads x head_size
TORCH_CHECK(k.stride(-1) == 1, "Input tensor must have contiguous last dimension");
TORCH_CHECK(v.stride(-1) == 1, "Input tensor must have contiguous last dimension");
TORCH_CHECK(q.is_contiguous(), "Input tensor must be contiguous");
TORCH_CHECK(k.is_contiguous(), "Input tensor must be contiguous");
TORCH_CHECK(v.is_contiguous(), "Input tensor must be contiguous");
const auto sizes = q.sizes();
const int batch_size = sizes[0];