CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

metal : tune soft_max number of threads (whisper/0)

Browse Source
This commit is contained in:
Georgi Gerganov
2024-05-13 11:01:07 +03:00
parent c3c88f296a
commit f308ea7059
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml-metal.m
View File

@@ -1378,7 +1378,7 @@ static enum ggml_status ggml_metal_graph_compute(
const bool use_f16 = (src1 && src1->type == GGML_TYPE_F16);
if (ne00%4 == 0) {
while (nth < ne00/4 && nth < 256) {
while (nth < ne00/4 && nth*ne01*ne02*ne03 < 256) {
nth *= 2;
}
if (use_f16) {
@@ -1387,7 +1387,7 @@ static enum ggml_status ggml_metal_graph_compute(
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32_4].pipeline;
}
} else {
while (nth < ne00 && nth < 1024) {
while (nth < ne00 && nth*ne01*ne02*ne03 < 256) {
nth *= 2;
}
if (use_f16) {