diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-06-05 19:13:51 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-06-05 19:13:51 +0300 |
| commit | eded4e20d4decdc6e8c18e645fd1db0833ad251d (patch) | |
| tree | b97f31f30d263b3d2ca610d75bcdd58793f7e04d /ggml/src/ggml-cuda/mmvq.cu | |
| parent | 8ffad187abbb93b74db8ef813b6fdceec80e02b0 (diff) | |
IQ1_M_R4 CUDA implementation (#494)
* iq1_m_r4: CUDA dequantize
* iq1_m_r4: CUDA dequantize
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml-cuda/mmvq.cu')
| -rw-r--r-- | ggml/src/ggml-cuda/mmvq.cu | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/mmvq.cu b/ggml/src/ggml-cuda/mmvq.cu index cc00d278..73caabab 100644 --- a/ggml/src/ggml-cuda/mmvq.cu +++ b/ggml/src/ggml-cuda/mmvq.cu @@ -563,6 +563,9 @@ static void ggml_cuda_op_mul_mat_vec_q_impl(ggml_backend_cuda_context & ctx, ggm case GGML_TYPE_IQ1_S_R4: mul_mat_vec_iq1_s_r4_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_M_R4: + mul_mat_vec_iq1_m_r4_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; default: GGML_ABORT("fatal error"); break; @@ -683,6 +686,7 @@ bool ggml_cuda_mmvq_type_supported(ggml_type src0_type) { case GGML_TYPE_IQ5_K_R4: case GGML_TYPE_IQ5_KS_R4: case GGML_TYPE_IQ1_S_R4: + case GGML_TYPE_IQ1_M_R4: return true; default: return false; |