diff --git a/ggml-metal.metal b/ggml-metal.metal index 0c6d32798..7af4e8f93 100644 --- a/ggml-metal.metal +++ b/ggml-metal.metal @@ -2217,7 +2217,7 @@ kernel void kernel_flash_attn_ext_f16( // ALiBi if (max_bias > 0.0f) { - const short h = iq2; + const uint32_t h = iq2; const float base = h < n_head_log2 ? m0 : m1; const int exph = h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1; @@ -2473,7 +2473,7 @@ kernel void kernel_flash_attn_ext_vec_f16( // ALiBi if (max_bias > 0.0f) { - const short h = iq2; + const uint32_t h = iq2; const float base = h < n_head_log2 ? m0 : m1; const int exp = h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1; diff --git a/scripts/sync-ggml.last b/scripts/sync-ggml.last index 46b874c64..1ea320429 100644 --- a/scripts/sync-ggml.last +++ b/scripts/sync-ggml.last @@ -1 +1 @@ -d2ad1793f45922665c99f5cf4244dcab77e16c00 +30f54cbb3ada3e4c5bc6924de3e5918e5be4ff11