vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

CUDA: fix FP16 cuBLAS GEMM (#11396)

Browse source
This commit is contained in:
Johannes Gäßler 2025-01-24 21:02:43 +01:00 • committed by GitHub
parent 9fbadaef4f
commit c5d9effb49
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

10
ggml/src/ggml-cuda/ggml-cuda.cu
View file

@ -1114,8 +1114,8 @@ static void ggml_cuda_op_mul_mat_cublas(
CUBLAS_CHECK( CUBLAS_CHECK(
cublasGemmEx(ctx.cublas_handle(id), CUBLAS_OP_T, CUBLAS_OP_N, cublasGemmEx(ctx.cublas_handle(id), CUBLAS_OP_T, CUBLAS_OP_N,
row_diff, src1_ncols, ne10, row_diff, src1_ncols, ne10,
&alpha, src0_ptr, CUDA_R_16F, ne00, &alpha, src0_ptr, CUDA_R_16F, ne00,
src1_ptr, CUDA_R_16F, ne10, src1_ptr, CUDA_R_16F, ne10,
&beta, dst_dd_i, CUDA_R_32F, ldc, &beta, dst_dd_i, CUDA_R_32F, ldc,
CUBLAS_COMPUTE_32F, CUBLAS_COMPUTE_32F,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); CUBLAS_GEMM_DEFAULT_TENSOR_OP));
@ -1128,9 +1128,9 @@ static void ggml_cuda_op_mul_mat_cublas(
CUBLAS_CHECK( CUBLAS_CHECK(
cublasGemmEx(ctx.cublas_handle(id), CUBLAS_OP_T, CUBLAS_OP_N, cublasGemmEx(ctx.cublas_handle(id), CUBLAS_OP_T, CUBLAS_OP_N,
row_diff, src1_ncols, ne10, row_diff, src1_ncols, ne10,
&alpha_f16, src0_ptr, CUDA_R_16F, ne00, &alpha_f16, src0_ptr, CUDA_R_16F, ne00,
src1_ptr, CUDA_R_16F, ne10, 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_COMPUTE_16F,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); CUBLAS_GEMM_DEFAULT_TENSOR_OP));