char-fork/llama.cpp

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

Johannes GäßlerCUDA: missing PDL sync for FWHT, better fallback (#23690)192d8ae8b

char/aliceai-t5-moe
192 B4 linesraw
1#include "common.cuh"
2
3// Returns whether the Fast Walsh-Hadamard transform could be used.
4bool ggml_cuda_op_fwht(ggml_backend_cuda_context & ctx, const ggml_tensor * src, ggml_tensor * dst);