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
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
|
- #!/usr/bin/env python3 -u
- # Copyright (c) 2017-present, Facebook, Inc.
- # All rights reserved.
- #
- # This source code is licensed under the license found in the LICENSE file in
- # the root directory of this source tree. An additional grant of patent rights
- # can be found in the PATENTS file in the same directory.
- """
- Translate raw text with a trained model. Batches data on-the-fly.
- """
- from collections import namedtuple
- import fileinput
- import sys
- import numpy as np
- import torch
- from fairseq import data, options, tasks, tokenizer, utils
- from fairseq.sequence_generator import SequenceGenerator
- from fairseq.utils import import_user_module
- Batch = namedtuple('Batch', 'srcs tokens lengths')
- Translation = namedtuple('Translation', 'src_str hypos pos_scores alignments')
- def buffered_read(input, buffer_size):
- buffer = []
- for src_str in fileinput.input(files=[input], openhook=fileinput.hook_encoded("utf-8")):
- buffer.append(src_str.strip())
- if len(buffer) >= buffer_size:
- yield buffer
- buffer = []
- if len(buffer) > 0:
- yield buffer
- def make_batches(lines, args, task, max_positions):
- tokens = [
- tokenizer.Tokenizer.tokenize(src_str, task.source_dictionary, add_if_not_exist=False).long()
- for src_str in lines
- ]
- lengths = np.array([t.numel() for t in tokens])
- itr = task.get_batch_iterator(
- dataset=data.LanguagePairDataset(tokens, lengths, task.source_dictionary),
- max_tokens=args.max_tokens,
- max_sentences=args.max_sentences,
- max_positions=max_positions,
- ).next_epoch_itr(shuffle=False)
- for batch in itr:
- yield Batch(
- srcs=[lines[i] for i in batch['id']],
- tokens=batch['net_input']['src_tokens'],
- lengths=batch['net_input']['src_lengths'],
- ), batch['id']
- def main(args):
- import_user_module(args)
- if args.buffer_size < 1:
- args.buffer_size = 1
- if args.max_tokens is None and args.max_sentences is None:
- args.max_sentences = 1
- assert not args.sampling or args.nbest == args.beam, \
- '--sampling requires --nbest to be equal to --beam'
- assert not args.max_sentences or args.max_sentences <= args.buffer_size, \
- '--max-sentences/--batch-size cannot be larger than --buffer-size'
- print(args)
- use_cuda = torch.cuda.is_available() and not args.cpu
- # Setup task, e.g., translation
- task = tasks.setup_task(args)
- # Load ensemble
- print('| loading model(s) from {}'.format(args.path))
- models, _model_args = utils.load_ensemble_for_inference(
- args.path.split(':'), task, model_arg_overrides=eval(args.model_overrides),
- )
- # Set dictionaries
- tgt_dict = task.target_dictionary
- # Optimize ensemble for generation
- for model in models:
- model.make_generation_fast_(
- beamable_mm_beam_size=None if args.no_beamable_mm else args.beam,
- need_attn=args.print_alignment,
- )
- if args.fp16:
- model.half()
- # Initialize generator
- translator = SequenceGenerator(
- models, tgt_dict, beam_size=args.beam, minlen=args.min_len,
- stop_early=(not args.no_early_stop), normalize_scores=(not args.unnormalized),
- len_penalty=args.lenpen, unk_penalty=args.unkpen,
- sampling=args.sampling, sampling_topk=args.sampling_topk, sampling_temperature=args.sampling_temperature,
- diverse_beam_groups=args.diverse_beam_groups, diverse_beam_strength=args.diverse_beam_strength,
- match_source_len=args.match_source_len, no_repeat_ngram_size=args.no_repeat_ngram_size,
- )
- if use_cuda:
- translator.cuda()
- # Load alignment dictionary for unknown word replacement
- # (None if no unknown word replacement, empty if no path to align dictionary)
- align_dict = utils.load_align_dict(args.replace_unk)
- def make_result(src_str, hypos):
- result = Translation(
- src_str='O\t{}'.format(src_str),
- hypos=[],
- pos_scores=[],
- alignments=[],
- )
- # Process top predictions
- for hypo in hypos[:min(len(hypos), args.nbest)]:
- hypo_tokens, hypo_str, alignment = utils.post_process_prediction(
- hypo_tokens=hypo['tokens'].int().cpu(),
- src_str=src_str,
- alignment=hypo['alignment'].int().cpu() if hypo['alignment'] is not None else None,
- align_dict=align_dict,
- tgt_dict=tgt_dict,
- remove_bpe=args.remove_bpe,
- )
- result.hypos.append('H\t{}\t{}'.format(hypo['score'], hypo_str))
- result.pos_scores.append('P\t{}'.format(
- ' '.join(map(
- lambda x: '{:.4f}'.format(x),
- hypo['positional_scores'].tolist(),
- ))
- ))
- result.alignments.append(
- 'A\t{}'.format(' '.join(map(lambda x: str(utils.item(x)), alignment)))
- if args.print_alignment else None
- )
- return result
- def process_batch(batch):
- tokens = batch.tokens
- lengths = batch.lengths
- if use_cuda:
- tokens = tokens.cuda()
- lengths = lengths.cuda()
- encoder_input = {'src_tokens': tokens, 'src_lengths': lengths}
- translations = translator.generate(
- encoder_input,
- maxlen=int(args.max_len_a * tokens.size(1) + args.max_len_b),
- )
- return [make_result(batch.srcs[i], t) for i, t in enumerate(translations)]
- max_positions = utils.resolve_max_positions(
- task.max_positions(),
- *[model.max_positions() for model in models]
- )
- if args.buffer_size > 1:
- print('| Sentence buffer size:', args.buffer_size)
- print('| Type the input sentence and press return:')
- for inputs in buffered_read(args.input, args.buffer_size):
- indices = []
- results = []
- for batch, batch_indices in make_batches(inputs, args, task, max_positions):
- indices.extend(batch_indices)
- results.extend(process_batch(batch))
- for i in np.argsort(indices):
- result = results[i]
- print(result.src_str)
- for hypo, pos_scores, align in zip(result.hypos, result.pos_scores, result.alignments):
- print(hypo)
- print(pos_scores)
- if align is not None:
- print(align)
- def cli_main():
- parser = options.get_generation_parser(interactive=True)
- args = options.parse_args_and_arch(parser)
- main(args)
- if __name__ == '__main__':
- cli_main()
|