llama.cpp/ggml-cuda/alibi.cuh

6 lines
134 B
Plaintext

#include "common.cuh"
#define CUDA_ALIBI_BLOCK_SIZE 32
void ggml_cuda_op_alibi(ggml_backend_cuda_context & ctx, ggml_tensor * dst);