ik-llama-cpp-sys 0.1.3

Low level bindings to ik_llama.cpp (ikawrakow's SOTA-quant fork)
Documentation 
1
2
3
4
5
6
7
8
9

#include "common.cuh"

void ggml_cuda_op_topk_moe(ggml_backend_cuda_context & ctx,
                           const ggml_tensor *         logits,
                           ggml_tensor *               weights,
                           ggml_tensor *               top_k,
                           ggml_tensor *               bias = nullptr);

bool ggml_cuda_should_use_topk_moe(const ggml_tensor * softmax, const ggml_tensor * weights);