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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
|
#include "lm/model.hh"
#include "lm/sizes.hh"
#include "util/file_piece.hh"
#include "util/usage.hh"
#include <algorithm>
#include <cstdlib>
#include <exception>
#include <iostream>
#include <iomanip>
#include <limits>
#include <math.h>
#include <stdlib.h>
#ifdef WIN32
#include "util/getopt.hh"
#else
#include <unistd.h>
#endif
namespace lm {
namespace ngram {
namespace {
void Usage(const char *name, const char *default_mem) {
std::cerr << "Usage: " << name << " [-u log10_unknown_probability] [-s] [-i] [-w mmap|after] [-p probing_multiplier] [-T trie_temporary] [-S trie_building_mem] [-q bits] [-b bits] [-a bits] [type] input.arpa [output.mmap]\n\n"
"-u sets the log10 probability for <unk> if the ARPA file does not have one.\n"
" Default is -100. The ARPA file will always take precedence.\n"
"-s allows models to be built even if they do not have <s> and </s>.\n"
"-i allows buggy models from IRSTLM by mapping positive log probability to 0.\n"
"-w mmap|after determines how writing is done.\n"
" mmap maps the binary file and writes to it. Default for trie.\n"
" after allocates anonymous memory, builds, and writes. Default for probing.\n"
"-r \"order1.arpa order2 order3 order4\" adds lower-order rest costs from these\n"
" model files. order1.arpa must be an ARPA file. All others may be ARPA or\n"
" the same data structure as being built. All files must have the same\n"
" vocabulary. For probing, the unigrams must be in the same order.\n\n"
"type is either probing or trie. Default is probing.\n\n"
"probing uses a probing hash table. It is the fastest but uses the most memory.\n"
"-p sets the space multiplier and must be >1.0. The default is 1.5.\n\n"
"trie is a straightforward trie with bit-level packing. It uses the least\n"
"memory and is still faster than SRI or IRST. Building the trie format uses an\n"
"on-disk sort to save memory.\n"
"-T is the temporary directory prefix. Default is the output file name.\n"
"-S determines memory use for sorting. Default is " << default_mem << ". This is compatible\n"
" with GNU sort. The number is followed by a unit: \% for percent of physical\n"
" memory, b for bytes, K for Kilobytes, M for megabytes, then G,T,P,E,Z,Y. \n"
" Default unit is K for Kilobytes.\n"
"-q turns quantization on and sets the number of bits (e.g. -q 8).\n"
"-b sets backoff quantization bits. Requires -q and defaults to that value.\n"
"-a compresses pointers using an array of offsets. The parameter is the\n"
" maximum number of bits encoded by the array. Memory is minimized subject\n"
" to the maximum, so pick 255 to minimize memory.\n\n"
"Get a memory estimate by passing an ARPA file without an output file name.\n";
exit(1);
}
// I could really use boost::lexical_cast right about now.
float ParseFloat(const char *from) {
char *end;
float ret = strtod(from, &end);
if (*end) throw util::ParseNumberException(from);
return ret;
}
unsigned long int ParseUInt(const char *from) {
char *end;
unsigned long int ret = strtoul(from, &end, 10);
if (*end) throw util::ParseNumberException(from);
return ret;
}
uint8_t ParseBitCount(const char *from) {
unsigned long val = ParseUInt(from);
if (val > 25) {
util::ParseNumberException e(from);
e << " bit counts are limited to 25.";
}
return val;
}
void ParseFileList(const char *from, std::vector<std::string> &to) {
to.clear();
while (true) {
const char *i;
for (i = from; *i && *i != ' '; ++i) {}
to.push_back(std::string(from, i - from));
if (!*i) break;
from = i + 1;
}
}
void ProbingQuantizationUnsupported() {
std::cerr << "Quantization is only implemented in the trie data structure." << std::endl;
exit(1);
}
} // namespace ngram
} // namespace lm
} // namespace
int main(int argc, char *argv[]) {
using namespace lm::ngram;
const char *default_mem = util::GuessPhysicalMemory() ? "80%" : "1G";
try {
bool quantize = false, set_backoff_bits = false, bhiksha = false, set_write_method = false, rest = false;
lm::ngram::Config config;
config.building_memory = util::ParseSize(default_mem);
int opt;
while ((opt = getopt(argc, argv, "q:b:a:u:p:t:T:m:S:w:sir:")) != -1) {
switch(opt) {
case 'q':
config.prob_bits = ParseBitCount(optarg);
if (!set_backoff_bits) config.backoff_bits = config.prob_bits;
quantize = true;
break;
case 'b':
config.backoff_bits = ParseBitCount(optarg);
set_backoff_bits = true;
break;
case 'a':
config.pointer_bhiksha_bits = ParseBitCount(optarg);
bhiksha = true;
break;
case 'u':
config.unknown_missing_logprob = ParseFloat(optarg);
break;
case 'p':
config.probing_multiplier = ParseFloat(optarg);
break;
case 't': // legacy
case 'T':
config.temporary_directory_prefix = optarg;
break;
case 'm': // legacy
config.building_memory = ParseUInt(optarg) * 1048576;
break;
case 'S':
config.building_memory = std::min(static_cast<uint64_t>(std::numeric_limits<std::size_t>::max()), util::ParseSize(optarg));
break;
case 'w':
set_write_method = true;
if (!strcmp(optarg, "mmap")) {
config.write_method = Config::WRITE_MMAP;
} else if (!strcmp(optarg, "after")) {
config.write_method = Config::WRITE_AFTER;
} else {
Usage(argv[0], default_mem);
}
break;
case 's':
config.sentence_marker_missing = lm::SILENT;
break;
case 'i':
config.positive_log_probability = lm::SILENT;
break;
case 'r':
rest = true;
ParseFileList(optarg, config.rest_lower_files);
config.rest_function = Config::REST_LOWER;
break;
default:
Usage(argv[0], default_mem);
}
}
if (!quantize && set_backoff_bits) {
std::cerr << "You specified backoff quantization (-b) but not probability quantization (-q)" << std::endl;
abort();
}
if (optind + 1 == argc) {
ShowSizes(argv[optind], config);
return 0;
}
const char *model_type;
const char *from_file;
if (optind + 2 == argc) {
model_type = "probing";
from_file = argv[optind];
config.write_mmap = argv[optind + 1];
} else if (optind + 3 == argc) {
model_type = argv[optind];
from_file = argv[optind + 1];
config.write_mmap = argv[optind + 2];
} else {
Usage(argv[0], default_mem);
}
if (!strcmp(model_type, "probing")) {
if (!set_write_method) config.write_method = Config::WRITE_AFTER;
if (quantize || set_backoff_bits) ProbingQuantizationUnsupported();
if (rest) {
RestProbingModel(from_file, config);
} else {
ProbingModel(from_file, config);
}
} else if (!strcmp(model_type, "trie")) {
if (rest) {
std::cerr << "Rest + trie is not supported yet." << std::endl;
return 1;
}
if (!set_write_method) config.write_method = Config::WRITE_MMAP;
if (quantize) {
if (bhiksha) {
QuantArrayTrieModel(from_file, config);
} else {
QuantTrieModel(from_file, config);
}
} else {
if (bhiksha) {
ArrayTrieModel(from_file, config);
} else {
TrieModel(from_file, config);
}
}
} else {
Usage(argv[0], default_mem);
}
}
catch (const std::exception &e) {
std::cerr << e.what() << std::endl;
std::cerr << "ERROR" << std::endl;
return 1;
}
std::cerr << "SUCCESS" << std::endl;
return 0;
}
|
