hf_text-generation-inference/server/marlin/marlin_kernels/ext.hh

24 lines
931 B
C++

#pragma once
#include <torch/library.h>
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
// No support for async
#else
torch::Tensor gptq_marlin_gemm(torch::Tensor &a, torch::Tensor &b_q_weight,
torch::Tensor &b_scales, torch::Tensor &g_idx,
torch::Tensor &perm, torch::Tensor &workspace,
int64_t num_bits, int64_t size_m, int64_t size_n,
int64_t size_k, bool is_k_full);
torch::Tensor gptq_marlin_repack(torch::Tensor &b_q_weight, torch::Tensor &perm,
int64_t size_k, int64_t size_n,
int64_t num_bits);
torch::Tensor marlin_gemm(torch::Tensor &a, torch::Tensor &b_q_weight,
torch::Tensor &b_scales, torch::Tensor &workspace,
int64_t size_m, int64_t size_n, int64_t size_k);
#endif