diff options
Diffstat (limited to 'ggml/src/ggml-cuda/mmq.cu')
| -rw-r--r-- | ggml/src/ggml-cuda/mmq.cu | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/mmq.cu b/ggml/src/ggml-cuda/mmq.cu index 60c2037f..a13be11b 100644 --- a/ggml/src/ggml-cuda/mmq.cu +++ b/ggml/src/ggml-cuda/mmq.cu @@ -97,9 +97,15 @@ void ggml_cuda_op_mul_mat_q( case GGML_TYPE_IQ4_KS: mul_mat_q_case<GGML_TYPE_IQ4_KS>(ctx, args, stream); break; + case GGML_TYPE_IQ4_KS_R4: + mul_mat_q_case<GGML_TYPE_IQ4_KS_R4>(ctx, args, stream); + break; case GGML_TYPE_IQ5_KS: mul_mat_q_case<GGML_TYPE_IQ5_KS>(ctx, args, stream); break; + case GGML_TYPE_IQ5_KS_R4: + mul_mat_q_case<GGML_TYPE_IQ5_KS_R4>(ctx, args, stream); + break; case GGML_TYPE_IQ2_KS: mul_mat_q_case<GGML_TYPE_IQ2_KS>(ctx, args, stream); break; @@ -157,7 +163,9 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11) { case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_KS: + case GGML_TYPE_IQ4_KS_R4: case GGML_TYPE_IQ5_KS: + case GGML_TYPE_IQ5_KS_R4: case GGML_TYPE_IQ2_KS: case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ3_K: |