zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-08-02 16:56:46 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Disable hd=128 until further optimizations

Browse Source
This commit is contained in:
Jagrit Digani 2024-11-21 11:21:38 -08:00
parent 990b1acc75
commit 0404037ea6
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
mlx/fast.cpp
View File

@ -644,7 +644,7 @@ array scaled_dot_product_attention(
const bool sdpa_vector_supported_head_dim = const bool sdpa_vector_supported_head_dim =
query_head_dim == 64 || query_head_dim == 96 || query_head_dim == 128; query_head_dim == 64 || query_head_dim == 96 || query_head_dim == 128;
const bool sdpa_full_supported_head_dim = const bool sdpa_full_supported_head_dim =
query_head_dim == 64 || query_head_dim == 80 || query_head_dim == 128; query_head_dim == 64 || query_head_dim == 80;
const bool supports_sdpa_full = query_sequence_length >= threshold && const bool supports_sdpa_full = query_sequence_length >= threshold &&
!mask.has_value() && sdpa_full_supported_head_dim && !mask.has_value() && sdpa_full_supported_head_dim &&