summaryrefslogtreecommitdiff
path: root/ggml/src/ggml-cuda/mmvq.cu
diff options
context:
space:
mode:
Diffstat (limited to 'ggml/src/ggml-cuda/mmvq.cu')
-rw-r--r--ggml/src/ggml-cuda/mmvq.cu4
1 files changed, 4 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/mmvq.cu b/ggml/src/ggml-cuda/mmvq.cu
index d0746031..012b3e5e 100644
--- a/ggml/src/ggml-cuda/mmvq.cu
+++ b/ggml/src/ggml-cuda/mmvq.cu
@@ -533,6 +533,9 @@ static void ggml_cuda_op_mul_mat_vec_q_impl(ggml_backend_cuda_context & ctx, ggm
case GGML_TYPE_IQ4_KSS:
mul_mat_vec_iq4_kss_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ids_data, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, ne2, nb02, nb12, nb2, ids_nb0, stream);
break;
+ case GGML_TYPE_IQ1_KT:
+ mul_mat_vec_iq1_kt_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ids_data, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, ne2, nb02, nb12, nb2, ids_nb0, stream);
+ break;
case GGML_TYPE_IQ2_KT:
mul_mat_vec_iq2_kt_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ids_data, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, ne2, nb02, nb12, nb2, ids_nb0, stream);
break;
@@ -704,6 +707,7 @@ bool ggml_cuda_mmvq_type_supported(ggml_type src0_type) {
case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ1_S_R4:
case GGML_TYPE_IQ1_M_R4:
+ case GGML_TYPE_IQ1_KT:
case GGML_TYPE_IQ2_KT:
case GGML_TYPE_IQ3_KT:
case GGML_TYPE_IQ4_KT: