File size: 1,262 Bytes
6fc683c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
/*
Copyright (c) Microsoft Corporation.
Licensed under the MIT License.
*/

#include <torch/extension.h>
#include <vector>

/*
CPP Binding for CUDA OP
*/

// CUDA forward declarations
torch::Tensor ngram_repeat_block_cuda_forward(
    torch::Tensor tokens,
    torch::Tensor lprobs,
    int bsz,
    int step,
    int beam_size,
    int no_repeat_ngram_size);

#define CHECK_CUDA(x) \
  TORCH_CHECK(x.type().is_cuda(), #x " must be a CUDA tensor")
#define CHECK_CONTIGUOUS(x) \
  TORCH_CHECK(x.is_contiguous(), #x " must be contiguous")
#define CHECK_INPUT(x) \
  CHECK_CUDA(x);       \
  CHECK_CONTIGUOUS(x)

// Input check and call to CUDA OP
// Backward method not required
torch::Tensor ngram_repeat_block_forward(
    torch::Tensor tokens,
    torch::Tensor lprobs,
    int bsz,
    int step,
    int beam_size,
    int no_repeat_ngram_size) {
  CHECK_INPUT(tokens);
  CHECK_INPUT(lprobs);
  assert(bsz > 0);
  assert(step >= 0);
  assert(beam_size > 0);
  assert(no_repeat_ngram_size > 0);

  return ngram_repeat_block_cuda_forward(
      tokens, lprobs, bsz, step, beam_size, no_repeat_ngram_size);
}

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
  m.def(
      "forward",
      &ngram_repeat_block_forward,
      "No Repeat Ngram Block forward (CUDA)");
}