happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix chunks being too small with small matrix sizes (#17526)

This commit is contained in:
Alberto Cabrera Pérez 2025-11-26 21:14:54 +00:00 committed by GitHub
parent 1d594c295c
commit 5449367b21
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-cpu/repack.cpp
View File

@ -1731,12 +1731,13 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
nchunk0 = (nr0 + min_chunk_size - 1) / min_chunk_size;
}
if (nth == 1 || nchunk0 < nth || disable_chunking) {
int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
// Only increase nchunk0 to nth if it won't make chunks too small
if (nth == 1 || ((nchunk0 < nth || disable_chunking) && (nr0 + nth - 1) / nth >= min_chunk_size)) {
nchunk0 = nth;
dr0 = (nr0 + nchunk0 - 1) / nchunk0;
}
const int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
// Ensure nchunk doesn't exceed the number of rows divided by minimum chunk size
// This prevents creating too many tiny chunks that could overlap after alignment
const int64_t max_nchunk = (nr0 + min_chunk_size - 1) / min_chunk_size;