diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-02-24 09:29:58 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-02-24 09:29:58 +0200 |
| commit | 547eee81d99a2676975a9768166b7d164473b8fa (patch) | |
| tree | cc248c7734ac9a7e16104767093170841038f158 | |
| parent | ac1d259b93eccfa7371c6b00c5749400ff2b2aea (diff) | |
Fix #230 (#231)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
| -rw-r--r-- | ggml/src/iqk/iqk_mul_mat.cpp | 7 |
1 files changed, 3 insertions, 4 deletions
diff --git a/ggml/src/iqk/iqk_mul_mat.cpp b/ggml/src/iqk/iqk_mul_mat.cpp index 0f7cd1e5..0955f15d 100644 --- a/ggml/src/iqk/iqk_mul_mat.cpp +++ b/ggml/src/iqk/iqk_mul_mat.cpp @@ -4062,7 +4062,7 @@ static void mul_mat_q4_0_r8_q8_1(int n, const void * vx, size_t bx, const DataIn template <int nrc_y> static void mul_mat_q5_0_r4_q8_1_avx2(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) { - GGML_ASSERT(nrc_x%8 == 0); + GGML_ASSERT(nrc_x%4 == 0); Q8<nrc_y, block_q8_1_x4> q8(info); auto m4 = _mm256_set1_epi8(0xf); auto m5 = _mm256_set1_epi8(0x10); @@ -4232,7 +4232,7 @@ static void mul_mat_q5_0_r4_q8_1(int n, const void * vx, size_t bx, const DataIn template <int nrc_y> static void mul_mat_q6_0_r4_q8_1_avx2(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) { - GGML_ASSERT(nrc_x%8 == 0); + GGML_ASSERT(nrc_x%4 == 0); Q8<nrc_y, block_q8_1_x4> q8(info); auto m4 = _mm256_set1_epi8(0xf); auto m6 = _mm256_set1_epi8(0x30); @@ -6493,7 +6493,6 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI #endif return; } - GGML_ASSERT(nrc_x%8 == 0); __m256i qx[2]; __m256i acc[2*nrc_y] = {}; float dy[nrc_y]; @@ -6566,7 +6565,7 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI template <int nrc_y> static void mul_mat_q8_KV_q8_KV(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) { - GGML_ASSERT(nrc_x%8 == 0); + GGML_ASSERT(nrc_x%4 == 0); GGML_ASSERT(n%32 == 0); __m256i qx[4]; #ifndef HAVE_FANCY_SIMD |