diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-06-14 05:52:46 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-06-14 05:52:46 +0300 |
| commit | 6fc5bbb657525bb1ef20b682e1cc4ab5fd44aba6 (patch) | |
| tree | 507baa41c53a097e704c9d3b917e7dcd50cb25e4 | |
| parent | 066ed4fd1158ddaab0080ef0e77bd5b7e12ec114 (diff) | |
Call iqk_convert_repack in MoE GEMM (#528)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
| -rw-r--r-- | ggml/src/iqk/iqk_mul_mat.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/ggml/src/iqk/iqk_mul_mat.cpp b/ggml/src/iqk/iqk_mul_mat.cpp index 7c0d3aff..b23dc6d4 100644 --- a/ggml/src/iqk/iqk_mul_mat.cpp +++ b/ggml/src/iqk/iqk_mul_mat.cpp @@ -626,7 +626,7 @@ extern "C" IQK_API bool iqk_mul_mat_moe(long Nx, long Ny, long ne00, int ne11, this_info.s += ix; int this_nrc_x = ix + k_x_step <= nrc_x ? k_x_step : nrc_x - ix; if (f.size() < row_size_qx*this_nrc_x) f.resize(row_size_qx*this_nrc_x); - if (!iqk_dequantize_ktquants(typeA, ne00, (const char *)A + (first_x + ix)*strideA, strideA, f.data(), ne00, this_nrc_x)) { + if (!iqk_convert_repack(typeA, ne00, (const char *)A + (first_x + ix)*strideA, strideA, f.data(), ne00, this_nrc_x)) { GGML_ABORT("Fatal error"); } mm.mul_mat_NxM(ne00, f.data(), row_size_qx, this_info, this_nrc_x, Ny); @@ -696,10 +696,10 @@ extern "C" IQK_API bool iqk_moe_fused_up_gate(long Nx, long Ny, long ne00, int n if (f.size() < 2*row_size_qx*this_nrc_x) f.resize(2*row_size_qx*this_nrc_x); auto Xu = f.data(); auto Xg = f.data() + row_size_qx*this_nrc_x; - if (!iqk_dequantize_ktquants(typeA, ne00, (const char *)Aup + (first_x + ix)*strideA, strideA, Xu, ne00, this_nrc_x)) { + if (!iqk_convert_repack(typeA, ne00, (const char *)Aup + (first_x + ix)*strideA, strideA, Xu, ne00, this_nrc_x)) { GGML_ABORT("Fatal error"); } - if (!iqk_dequantize_ktquants(typeA, ne00, (const char *)Agate + (first_x + ix)*strideA, strideA, Xg, ne00, this_nrc_x)) { + if (!iqk_convert_repack(typeA, ne00, (const char *)Agate + (first_x + ix)*strideA, strideA, Xg, ne00, this_nrc_x)) { GGML_ABORT("Fatal error"); } mm.mul_mat_up_gate_NxM(ne00, Xu, Xg, row_size_qx, this_info, this_nrc_x, Ny, unary_op); |