CUDA: fix FP16 cuBLAS GEMM (#11396)

2025-01-24 21:02:43 +01:00 · 2025-01-24 21:02:43 +01:00 · c5d9effb49
commit c5d9effb49
parent 9fbadaef4f
1 changed files with 5 additions and 5 deletions
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@ -1130,7 +1130,7 @@ static void ggml_cuda_op_mul_mat_cublas(
                        row_diff, src1_ncols, ne10,
                        &alpha_f16, src0_ptr,      CUDA_R_16F, ne00,
                                    src1_ptr,      CUDA_R_16F, ne10,
-                        &beta_f16,   dst_dd_i, CUDA_R_16F, ldc,
+                        &beta_f16,  dst_f16.get(), CUDA_R_16F, ldc,
                        CUBLAS_COMPUTE_16F,
                        CUBLAS_GEMM_DEFAULT_TENSOR_OP));