aphrodite-engine/kernels/quantization/gptq/alt_matmul.cpp

#include <torch/all.h>
#include <torch/python.h>
#include <c10/cuda/CUDAGuard.h>

void vecquant4matmul_cuda(
    torch::Tensor vec,
    torch::Tensor mat,
    torch::Tensor mul,
    torch::Tensor scales,
    torch::Tensor zeros,
    torch::Tensor g_idx
);

void gptq_descact_matmul(
  torch::Tensor vec,
  torch::Tensor mat,
  torch::Tensor mul,
  torch::Tensor scales,
  torch::Tensor zeros,
  torch::Tensor g_idx
)
{
  const at::cuda::OptionalCUDAGuard device_guard(device_of(vec));
  vecquant4matmul_cuda(vec, mat, mul, scales, zeros, g_idx);
}