-
Notifications
You must be signed in to change notification settings - Fork 2
/
translate.py
executable file
·59 lines (44 loc) · 2.63 KB
/
translate.py
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
import fairseq,torch,sys,argparse
# Based on https://github.com/pytorch/fairseq/blob/master/examples/translation/README.md
# Install: pip install fairseq fastBPE sacremoses subword_nmt
# Usage: echo 'Hello World' | python translate.py -m transformer.wmt19.en-de
help_msg='See https://github.com/pytorch/fairseq/blob/master/examples/translation/README.md for models that are likely to work.\nModels will download (if necessary).\nIf you provide an unknown model, you will see a list of known models (though not all of them will work).'
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--model_string", help=help_msg, default=None)
parser.add_argument("-s", "--source_language", help="two letter language code such as en, de, zh, etc.", default=None)
parser.add_argument("-t", "--target_language", help="two letter language code such as en, de, zh, etc.", default=None)
args = parser.parse_args()
tested_model_strings = [
'transformer.wmt14.en-fr',
'transformer.wmt16.en-de',
'transformer.wmt19.de-en',
'transformer.wmt19.de-en.single_model',
'transformer.wmt19.en-de',
'transformer.wmt19.en-de.single_model',
'transformer.wmt19.en-ru',
'transformer.wmt19.en-ru.single_model',
'transformer.wmt19.ru-en',
'transformer.wmt19.ru-en.single_model']
if args.model_string is None:
args.model_string = 'transformer.wmt19.' + args.source_language + '-' + args.target_language + '.single_model'
print('model_string: ' + str(args.model_string))
model_strings = torch.hub.list('pytorch/fairseq') # Our code is known to fail on some of these
assert args.model_string in model_strings, help_msg + '\n%s should be one of the following (though not all of them will work): %s' % (args.model_string, '\n\t'.join(model_strings))
if not args.model_string in tested_model_strings:
print('warning: %s may not work; the following have been tested: %s' % (args.model_string, '\n\t'.join(tested_model_strings)), file=sys.stderr)
# Warning, there are some nasty interactions between models and arguments
bpe='subword_nmt'
checkpoint_file='model.pt'
if args.model_string == 'transformer.wmt18.en-de':
checkpoint_file='nc_model/model.pt'
if args.model_string.find('wmt19') >= 0:
bpe='fastbpe'
if not args.model_string.endswith('single_model'):
checkpoint_file='model1.pt:model2.pt:model3.pt:model4.pt'
model = torch.hub.load('pytorch/fairseq', args.model_string, tokenizer='moses', bpe=bpe, checkpoint_file=checkpoint_file)
model.cuda()
model.eval()
assert isinstance(model.models[0], fairseq.models.transformer.TransformerModel)
for line in sys.stdin:
sent = line.rstrip()
print(sent + '\t' + model.translate(sent))