From 46dba9fce860d41ac545224623f27ac71f9d264a Mon Sep 17 00:00:00 2001 From: Ruben Ortlam Date: Mon, 16 Mar 2026 10:45:49 +0100 Subject: [PATCH] vulkan: fix flash attention dot product precision (#20589) --- ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp index ec48f5b115..11b7dce857 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp @@ -245,7 +245,7 @@ void main() { #endif } [[unroll]] for (uint32_t r = 0; r < rows_per_thread; ++r) { - Sf[r][c] += ACC_TYPE(dot(Q_cache[r], K_Tf)); + Sf[r][c] += dot(ACC_TYPEV4(Q_cache[r]), ACC_TYPEV4(K_Tf)); } } } @@ -270,7 +270,7 @@ void main() { #endif } [[unroll]] for (uint32_t r = 0; r < rows_per_thread; ++r) { - Sf[r][c] += ACC_TYPE(dot(Qf[tile_row(r) * qf_stride + d * D_split + d_tid], K_Tf)); + Sf[r][c] += dot(ACC_TYPEV4(Qf[tile_row(r) * qf_stride + d * D_split + d_tid]), ACC_TYPEV4(K_Tf)); } } }