From 44fc0a5eb6208f31205fd4eb5f8f4bab7a9c467a Mon Sep 17 00:00:00 2001 From: Iwan Kawrakow Date: Wed, 27 Aug 2025 14:24:35 +0300 Subject: [PATCH] sanitize imatrix: iq5_ks --- ggml/src/iqk/iqk_quantize.cpp | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/ggml/src/iqk/iqk_quantize.cpp b/ggml/src/iqk/iqk_quantize.cpp index e22edba1..dd7d192e 100644 --- a/ggml/src/iqk/iqk_quantize.cpp +++ b/ggml/src/iqk/iqk_quantize.cpp @@ -4289,14 +4289,13 @@ size_t quantize_iq5_ks(const float * src, void * dst, int64_t nrows, int64_t n_p constexpr int kBlockSize = 32; GGML_ASSERT(n_per_row%QK_K == 0); auto row_size = ggml_row_size(GGML_TYPE_IQ5_KS, n_per_row); - char * qrow = (char *)dst; float weight[kBlockSize]; std::vector all_scales(n_per_row/kBlockSize); - for (int64_t row = 0; row < nrows; ++row) { - quantize_row_iq5_ks_impl(QK_K, kBlockSize, n_per_row, src, qrow, all_scales.data(), weight, iq5nl_values, imatrix, 5); - src += n_per_row; - qrow += row_size; - } + QHelper helper(imatrix, n_per_row, kBlockSize); + auto q_func = [&all_scales, &weight, block_size = kBlockSize] (const float * x, void * vy, int n_per_row, const float * imatrix) { + quantize_row_iq5_ks_impl(QK_K, block_size, n_per_row, x, (char *)vy, all_scales.data(), weight, iq5nl_values, imatrix, 5); + }; + helper.quantize(nrows, src, dst, row_size, q_func); return nrows * row_size; }