char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Sigbjørn Skjæretcuda : add softcap fusion (#14907)138b288b5

char/aliceai-t5-moe
158 B5 linesraw
1#include "common.cuh"
2
3#define CUDA_SOFTCAP_BLOCK_SIZE 256
4
5void ggml_cuda_op_softcap(ggml_backend_cuda_context & ctx, ggml_tensor * dst, ggml_tensor * src);