From 37d72f9878ed0ee4b55fcb95c8778706394d83f4 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Sun, 16 Nov 2025 11:50:29 +0200 Subject: [PATCH] Fix ggml_cuda_fattn_is_supported (#968) Co-authored-by: Iwan Kawrakow --- ggml/src/ggml-cuda/fattn.cu | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ggml/src/ggml-cuda/fattn.cu b/ggml/src/ggml-cuda/fattn.cu index 1ed10bd9..73a7c3bb 100644 --- a/ggml/src/ggml-cuda/fattn.cu +++ b/ggml/src/ggml-cuda/fattn.cu @@ -169,7 +169,7 @@ bool ggml_cuda_fattn_is_supported(ggml_backend_cuda_context & ctx, const ggml_te const bool mma_faster_for_bs1 = new_mma_available(cc) && gqa_opt_applies && !(Q->ne[1] == 1 && n_swa > 0); const bool can_use_vector_kernel = Q->ne[0] <= 256 && Q->ne[0] % (2*WARP_SIZE) == 0; if (Q->ne[1] == 1 && can_use_vector_kernel && !mma_faster_for_bs1 && !ggml_is_quantized(K->type) && !ggml_is_quantized(V->type)) { - return ggml_cuda_fattn_tile_f32_is_supported(ctx, dst); + return ggml_cuda_fattn_vec_f32_is_supported(ctx, dst); } if (new_mma_available(cc) && (Q->ne[0] == 576 || (K->ne[0] == 192 && V->ne[0] == 128))) {