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
|
// Copyright 2016 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.!
#include <map>
#include "flags.h"
#include "sentencepiece_model.pb.h"
#include "sentencepiece_trainer.h"
#include "util.h"
using sentencepiece::NormalizerSpec;
using sentencepiece::TrainerSpec;
namespace {
static sentencepiece::TrainerSpec kDefaultTrainerSpec;
static sentencepiece::NormalizerSpec kDefaultNormalizerSpec;
} // namespace
DEFINE_string(input, "", "comma separated list of input sentences");
DEFINE_string(input_format, kDefaultTrainerSpec.input_format(),
"Input format. Supported format is `text` or `tsv`.");
DEFINE_string(model_prefix, "", "output model prefix");
DEFINE_string(model_type, "unigram",
"model algorithm: unigram, bpe, word or char");
DEFINE_int32(vocab_size, kDefaultTrainerSpec.vocab_size(), "vocabulary size");
DEFINE_string(accept_language, "",
"comma-separated list of languages this model can accept");
DEFINE_double(character_coverage, kDefaultTrainerSpec.character_coverage(),
"character coverage to determine the minimum symbols");
DEFINE_int32(input_sentence_size, kDefaultTrainerSpec.input_sentence_size(),
"maximum size of sentences the trainer loads");
DEFINE_int32(mining_sentence_size, kDefaultTrainerSpec.mining_sentence_size(),
"maximum size of sentences to make seed sentence piece");
DEFINE_int32(training_sentence_size,
kDefaultTrainerSpec.training_sentence_size(),
"maximum size of sentences to train sentence pieces");
DEFINE_int32(seed_sentencepiece_size,
kDefaultTrainerSpec.seed_sentencepiece_size(),
"the size of seed sentencepieces");
DEFINE_double(shrinking_factor, kDefaultTrainerSpec.shrinking_factor(),
"Keeps top shrinking_factor pieces with respect to the loss");
DEFINE_int32(num_threads, kDefaultTrainerSpec.num_threads(),
"number of threads for training");
DEFINE_int32(num_sub_iterations, kDefaultTrainerSpec.num_sub_iterations(),
"number of EM sub-iterations");
DEFINE_int32(max_sentencepiece_length,
kDefaultTrainerSpec.max_sentencepiece_length(),
"maximum length of sentence piece");
DEFINE_bool(split_by_unicode_script,
kDefaultTrainerSpec.split_by_unicode_script(),
"use Unicode script to split sentence pieces");
DEFINE_bool(split_by_whitespace, kDefaultTrainerSpec.split_by_whitespace(),
"use a white space to split sentence pieces");
DEFINE_string(control_symbols, "", "comma separated list of control symbols");
DEFINE_string(user_defined_symbols, "",
"comma separated list of user defined symbols");
DEFINE_string(normalization_rule_name, "nmt_nfkc",
"Normalization rule name. "
"Choose from nfkc or identity");
DEFINE_string(normalization_rule_tsv, "", "Normalization rule TSV file. ");
DEFINE_bool(add_dummy_prefix, kDefaultNormalizerSpec.add_dummy_prefix(),
"Add dummy whitespace at the beginning of text");
DEFINE_bool(remove_extra_whitespaces,
kDefaultNormalizerSpec.remove_extra_whitespaces(),
"Removes leading, trailing, and "
"duplicate internal whitespace");
DEFINE_bool(hard_vocab_limit, kDefaultTrainerSpec.hard_vocab_limit(),
"If set to false, --vocab_size is considered as a soft limit.");
DEFINE_int32(unk_id, kDefaultTrainerSpec.unk_id(), "Override UNK (<unk>) id.");
DEFINE_int32(bos_id, kDefaultTrainerSpec.bos_id(),
"Override BOS (<s>) id. Set -1 to disable BOS.");
DEFINE_int32(eos_id, kDefaultTrainerSpec.eos_id(),
"Override EOS (</s>) id. Set -1 to disable EOS.");
DEFINE_int32(pad_id, kDefaultTrainerSpec.pad_id(),
"Override PAD (<pad>) id. Set -1 to disable PAD.");
DEFINE_string(unk_surface, kDefaultTrainerSpec.unk_surface(),
"Dummy surface string for <unk>. In decoding <unk> is decoded to "
"`unk_surface`.");
int main(int argc, char *argv[]) {
sentencepiece::flags::ParseCommandLineFlags(argc, argv);
sentencepiece::TrainerSpec trainer_spec;
sentencepiece::NormalizerSpec normalizer_spec;
CHECK_OR_HELP(input);
CHECK_OR_HELP(model_prefix);
// Populates the value from flags to spec.
#define SetTrainerSpecFromFlag(name) trainer_spec.set_##name(FLAGS_##name);
#define SetNormalizerSpecFromFlag(name) \
normalizer_spec.set_##name(FLAGS_##name);
#define SetRepeatedTrainerSpecFromFlag(name) \
if (!FLAGS_##name.empty()) { \
for (const auto v : \
sentencepiece::string_util::Split(FLAGS_##name, ",")) { \
trainer_spec.add_##name(v); \
} \
}
SetTrainerSpecFromFlag(input_format);
SetTrainerSpecFromFlag(model_prefix);
SetTrainerSpecFromFlag(vocab_size);
SetTrainerSpecFromFlag(character_coverage);
SetTrainerSpecFromFlag(input_sentence_size);
SetTrainerSpecFromFlag(mining_sentence_size);
SetTrainerSpecFromFlag(training_sentence_size);
SetTrainerSpecFromFlag(seed_sentencepiece_size);
SetTrainerSpecFromFlag(shrinking_factor);
SetTrainerSpecFromFlag(num_threads);
SetTrainerSpecFromFlag(num_sub_iterations);
SetTrainerSpecFromFlag(max_sentencepiece_length);
SetTrainerSpecFromFlag(split_by_unicode_script);
SetTrainerSpecFromFlag(split_by_whitespace);
SetTrainerSpecFromFlag(hard_vocab_limit);
SetTrainerSpecFromFlag(unk_id);
SetTrainerSpecFromFlag(bos_id);
SetTrainerSpecFromFlag(eos_id);
SetTrainerSpecFromFlag(pad_id);
SetTrainerSpecFromFlag(unk_surface);
SetRepeatedTrainerSpecFromFlag(input);
SetRepeatedTrainerSpecFromFlag(accept_language);
SetRepeatedTrainerSpecFromFlag(control_symbols);
SetRepeatedTrainerSpecFromFlag(user_defined_symbols);
normalizer_spec.set_name(FLAGS_normalization_rule_name);
SetNormalizerSpecFromFlag(normalization_rule_tsv);
SetNormalizerSpecFromFlag(add_dummy_prefix);
SetNormalizerSpecFromFlag(remove_extra_whitespaces);
const std::map<std::string, TrainerSpec::ModelType> kModelTypeMap = {
{"unigram", TrainerSpec::UNIGRAM},
{"bpe", TrainerSpec::BPE},
{"word", TrainerSpec::WORD},
{"char", TrainerSpec::CHAR}};
trainer_spec.set_model_type(sentencepiece::port::FindOrDie(
kModelTypeMap, sentencepiece::string_util::ToLower(FLAGS_model_type)));
CHECK_OK(sentencepiece::SentencePieceTrainer::Train(trainer_spec,
normalizer_spec));
return 0;
}
|
