public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-10-18 04:35:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

metal : adjust extra size for FA buffer to avoid reallocations (llama/18545)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-01-14 09:11:59 +02:00
parent f328b13d5c
commit b99c911c49
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal-ops.cpp
+5 -1
View File
@@ -2181,7 +2181,11 @@ size_t ggml_metal_op_flash_attn_ext_extra_pad(const ggml_tensor * op) {
const bool has_mask = op->src[3] != nullptr;
if (ggml_metal_op_flash_attn_ext_use_vec(op)) {
// note: the non-vec kernel requires more extra memory, so always reserve for it
GGML_ASSERT(OP_FLASH_ATTN_EXT_NCPSG >= OP_FLASH_ATTN_EXT_VEC_NCPSG);
//if (ggml_metal_op_flash_attn_ext_use_vec(op)) {
if (false) {
// note: always reserve the padding space to avoid graph reallocations
//const bool has_kvpad = ne11 % OP_FLASH_ATTN_EXT_VEC_NCPSG != 0;
const bool has_kvpad = true;