Merge branch 'master' into gg/flash-attn

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-04-24 14:00:32 +03:00
commit 8937ec5307
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
57 changed files with 1576 additions and 8245 deletions

8
ggml.c
View file

@ -10948,7 +10948,7 @@ static void ggml_compute_forward_mul_mat(
#endif
#if GGML_USE_LLAMAFILE
if (nb10 == ggml_type_size(src1->type)) {
if (src1_cont) {
for (int64_t i13 = 0; i13 < ne13; i13++)
for (int64_t i12 = 0; i12 < ne12; i12++)
if (!llamafile_sgemm(ne01, ne11, ne00/ggml_blck_size(src0->type),
@ -11001,15 +11001,13 @@ UseGgmlGemm1:;
const size_t row_size = ggml_row_size(vec_dot_type, ne10);
#if GGML_USE_LLAMAFILE
if (nb10 == ggml_type_size(src1->type) || src1->type != vec_dot_type) {
if (src1->type != vec_dot_type) {
for (int64_t i13 = 0; i13 < ne13; i13++)
for (int64_t i12 = 0; i12 < ne12; i12++)
if (!llamafile_sgemm(ne01, ne11, ne00/ggml_blck_size(src0->type),
(const char *)src0->data + i12/r2*nb02 + i13/r3*nb03,
nb01/ggml_type_size(src0->type),
(const char *)wdata + ggml_row_size(vec_dot_type,
nb12/ggml_type_size(src1->type)*i12 +
nb13/ggml_type_size(src1->type)*i13),
(const char *)wdata + (i12*ne11 + i13*ne12*ne11)*row_size,
row_size/ggml_type_size(vec_dot_type),
(char *)dst->data + i12*nb2 + i13*nb3,
nb1/ggml_type_size(dst->type),