diff options
Diffstat (limited to 'ggml/src/ggml-cuda/iqk_mmvq.cuh')
| -rw-r--r-- | ggml/src/ggml-cuda/iqk_mmvq.cuh | 5 |
1 files changed, 5 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/iqk_mmvq.cuh b/ggml/src/ggml-cuda/iqk_mmvq.cuh index d14c3541..5d62d02e 100644 --- a/ggml/src/ggml-cuda/iqk_mmvq.cuh +++ b/ggml/src/ggml-cuda/iqk_mmvq.cuh @@ -111,6 +111,11 @@ void mul_mat_vec_iq1_m_r4_q8_1_cuda( const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, const int ne2, const uint64_t nb02, const uint64_t nb12, const uint64_t nb2, const int64_t ids_nb0, cudaStream_t stream); +void mul_mat_vec_iq1_kt_q8_1_cuda( + const void * vx, const void * vy, float * dst, const char * ids_data, + const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, + const int ne2, const uint64_t nb02, const uint64_t nb12, const uint64_t nb2, const int64_t ids_nb0, cudaStream_t stream); + void mul_mat_vec_iq2_kt_q8_1_cuda( const void * vx, const void * vy, float * dst, const char * ids_data, const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, |