char-fork/llama.cpp

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

Pascalcuda: fuse snake activation (mul, sin, sqr, mul, add) (#22667)58e68df0f

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
366 B8 linesraw
1#include "common.cuh"
2
3// Fusion entry point. Caller supplies x/a/inv_b explicitly.
4void ggml_cuda_op_snake_fused(ggml_backend_cuda_context & ctx,
5                              const ggml_tensor * x,
6                              const ggml_tensor * a,
7                              const ggml_tensor * inv_b,
8                              ggml_tensor *       dst);