From d18523c8e93e36b00df8a64090129495d473f182 Mon Sep 17 00:00:00 2001 From: Iwan Kawrakow Date: Wed, 12 Nov 2025 17:06:50 +0200 Subject: [PATCH] One more --- ggml/src/ggml-cuda/rope.cu | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml-cuda/rope.cu b/ggml/src/ggml-cuda/rope.cu index 113d1544..b513c876 100644 --- a/ggml/src/ggml-cuda/rope.cu +++ b/ggml/src/ggml-cuda/rope.cu @@ -508,7 +508,7 @@ static void rope_neox_fast_cuda(const float * src0, const float * src1, float * const dim3 block_dims(CUDA_ROPE_BLOCK_SIZE, 1, 1); const int n_blocks = (ne00*ne01*ne02 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE); const dim3 block_nums(n_blocks, 1, 1); - rope_neox_fast<<>>(src0, src1, dst, ne00, ne01, ne01*ne02*ne02, s01, s02, n_dims); + rope_neox_fast<<>>(src0, src1, dst, ne00, ne01, ne00*ne01*ne02, s01, s02, n_dims); } static void fused_rope_neox_fast_cuda(const float * src0_1, const float * src0_2, const float * src1, @@ -557,7 +557,7 @@ static void rope_norm_fast_cuda(const float * src0, const float * src1, float * const dim3 block_dims(CUDA_ROPE_BLOCK_SIZE, 1, 1); const int n_blocks = (ne00*ne01*ne02 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE); const dim3 block_nums(n_blocks, 1, 1); - rope_norm_fast<<>>(src0, src1, dst, ne00, ne01, ne01*ne02*ne02, s01, s02, n_dims); + rope_norm_fast<<>>(src0, src1, dst, ne00, ne01, ne00*ne01*ne02, s01, s02, n_dims); } static void rope_multi_fast_cuda(const float * src0, const float * src1, float * dst, int ne00, int ne01, int ne02, int s01, int s02,