diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-04-26 09:19:43 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-04-26 09:19:43 +0200 |
| commit | 9e846f0eb196ed543cb29753bfd6a21a936a5138 (patch) | |
| tree | c341f5a2037c8a8f92cc0f59d2980f0510372e83 /ggml/src | |
| parent | 715fc552ad2ea5fad38e7ff856bf84fdb71b692e (diff) | |
Fix division by zero bug (#349)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src')
| -rw-r--r-- | ggml/src/ggml.c | 2 | ||||
| -rw-r--r-- | ggml/src/iqk/iqk_flash_attn.cpp | 2 |
2 files changed, 2 insertions, 2 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index 88013f74..f3cfd9a0 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -21790,7 +21790,7 @@ struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threa const struct ggml_tensor * k = node->src[1]; if (q->ne[1] == 1 && q->ne[3] == 1 && q->ne[2]/k->ne[2] > 1 && n_tasks > 1 && k->ne[1]/32 > 1) { if (k->ne[2] > 1) { - int nk = 32 * (k->ne[2]*k->ne[1]/(32*n_tasks)); + int nk = MAX(1, 32 * (k->ne[2]*k->ne[1]/(32*n_tasks))); int nstep_k = k->ne[2]*k->ne[1]/nk; size_t result_size = (Dv + 16)*q->ne[2]/k->ne[2]*sizeof(float); size_t size = nstep_k*result_size; diff --git a/ggml/src/iqk/iqk_flash_attn.cpp b/ggml/src/iqk/iqk_flash_attn.cpp index 639b79af..0de68b94 100644 --- a/ggml/src/iqk/iqk_flash_attn.cpp +++ b/ggml/src/iqk/iqk_flash_attn.cpp @@ -154,7 +154,7 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float } if (neq3 == 1 && rk2 > 1 && rk2 == rv2 && neq1 == 1 && nth >= 1 && nek2*nek1 >= 32*nth) { - int nk = 32 * (nek2*nek1/(32*nth)); + int nk = std::max(1, 32 * (nek2*nek1/(32*nth))); int nkk = (nek1 + nk - 1)/nk; int nstep_k = nek2*nkk; auto result_size = (Dv + 16)*rk2*sizeof(float); |