12 lines
516 B
Plaintext
12 lines
516 B
Plaintext
#include <ATen/cuda/CUDAContext.h>
|
|
#include <torch/all.h>
|
|
|
|
#include "quantization/w8a8/per_token_group_quant_8bit.h"
|
|
|
|
void per_token_group_quant_int8(const torch::Tensor& input,
|
|
torch::Tensor& output_q,
|
|
torch::Tensor& output_s, int64_t group_size,
|
|
double eps, double int8_min, double int8_max) {
|
|
per_token_group_quant_8bit(input, output_q, output_s, group_size, eps,
|
|
int8_min, int8_max);
|
|
} |