and not pretrained_bert_parameters,
ctx=ctx,
use_pooler=False,
use_decoder=False,
use_classifier=False)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
berttoken = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
net = BertForQA(bert=bert)
if pretrained_bert_parameters and not model_parameters:
bert.load_parameters(pretrained_bert_parameters,
ctx=ctx,
ignore_extra=True)
if not model_parameters:
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
else:
net.load_parameters(model_parameters, ctx=ctx)
net.hybridize(static_alloc=True)
loss_function = BertForQALoss()
loss_function.hybridize(static_alloc=True)
def train():
if args.sentencepiece:
tokenizer = nlp.data.BERTSPTokenizer(args.sentencepiece,
vocab,
lower=lower)
else:
tokenizer = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
net = BertForQA(bert=bert)
if model_parameters:
# load complete BertForQA parameters
nlp.utils.load_parameters(net, model_parameters, ctx=ctx, cast_dtype=True)
elif pretrained_bert_parameters:
# only load BertModel parameters
nlp.utils.load_parameters(bert,
pretrained_bert_parameters,
ctx=ctx,
ignore_extra=True,
cast_dtype=True)
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
elif pretrained:
# only load BertModel parameters
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
else:
dataset_name=args.dataset_name,
pretrained=False,
use_pooler=True,
use_decoder=False,
use_classifier=False,
seq_length=args.seq_length)
net = BERTRegression(bert, dropout=args.dropout)
elif args.task == 'question_answering':
bert, _ = get_model(name=args.model_name,
dataset_name=args.dataset_name,
pretrained=False,
use_pooler=False,
use_decoder=False,
use_classifier=False,
seq_length=args.seq_length)
net = BertForQA(bert)
else:
raise ValueError('unknown task: %s' % args.task)
if args.model_parameters:
net.load_parameters(args.model_parameters)
else:
net.initialize()
warnings.warn(
'--model_parameters is not provided. The parameter checkpoint (.params) '
'file will be created based on default parameter initialization.')
net.hybridize(static_alloc=True, static_shape=True)
###############################################################################
# Prepare dummy input data #
lower=lower)
else:
tokenizer = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
BERT_DIM = {'bert_12_768_12': 768, 'bert_24_1024_16': 1024}
net = BertForQA(bert=bert, \
n_rnn_layers = args.n_rnn_layers,
apply_coattention=args.apply_coattention, bert_out_dim=BERT_DIM[args.bert_model], \
remove_special_token=args.remove_special_token,
mask_output=args.mask_output)
if model_parameters:
# load complete BertForQA parameters
net.load_parameters(model_parameters, ctx=ctx, cast_dtype=True)
elif pretrained_bert_parameters:
# only load BertModel parameters
bert.load_parameters(pretrained_bert_parameters,
ctx=ctx,
ignore_extra=True,
cast_dtype=True)
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
elif pretrained:
# only load BertModel parameters
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
tokenizer = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
BERT_DIM = {'bert_12_768_12': 768, 'bert_24_1024_16': 1024}
net = BertForQA(bert=bert, \
add_query=args.add_query, \
apply_coattention=args.apply_coattention, bert_out_dim=BERT_DIM[args.bert_model],\
apply_self_attention=args.apply_self_attention,
apply_transformer=args.apply_transformer,
qanet_style_out=args.qanet_style_out,
bidaf_style_out=args.bidaf_style_out,
n_rnn_layers=args.n_rnn_layers,
remove_special_token=args.remove_special_token)
# print(net)
# exit(0)
if args.apply_coattention and (args.qanet_style_out or args.bidaf_style_out):
additional_params = None
else:
additional_params = net.span_classifier.collect_params()
if model_parameters:
# load complete BertForQA parameters
net.load_parameters(model_parameters, ctx=ctx, cast_dtype=True)
vocab,
lower=lower)
else:
tokenizer = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
BERT_DIM = {'bert_12_768_12': 768, 'bert_24_1024_16': 1024}
net = BertForQA(bert=bert, \
add_query=args.add_query, \
apply_coattention=args.apply_coattention, bert_out_dim=BERT_DIM[args.bert_model],\
apply_self_attention=args.apply_self_attention)
if model_parameters:
# load complete BertForQA parameters
net.load_parameters(model_parameters, ctx=ctx, cast_dtype=True)
elif pretrained_bert_parameters:
# only load BertModel parameters
bert.load_parameters(pretrained_bert_parameters,
ctx=ctx,
ignore_extra=True,
cast_dtype=True)
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
elif pretrained:
# only load BertModel parameters
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
else:
if args.sentencepiece:
tokenizer = nlp.data.BERTSPTokenizer(args.sentencepiece,
vocab,
lower=lower)
else:
tokenizer = nlp.data.BERTTokenizer(vocab=vocab, lower=lower)
batchify_fn = nlp.data.batchify.Tuple(
nlp.data.batchify.Stack(),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Pad(axis=0, pad_val=vocab[vocab.padding_token]),
nlp.data.batchify.Stack('float32'), nlp.data.batchify.Stack('float32'),
nlp.data.batchify.Stack('float32'))
net = BertForQA(bert=bert)
if model_parameters:
# load complete BertForQA parameters
net.load_parameters(model_parameters, ctx=ctx, cast_dtype=True)
elif pretrained_bert_parameters:
# only load BertModel parameters
bert.load_parameters(pretrained_bert_parameters,
ctx=ctx,
ignore_extra=True,
cast_dtype=True)
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
elif pretrained:
# only load BertModel parameters
net.span_classifier.initialize(init=mx.init.Normal(0.02), ctx=ctx)
else:
# no checkpoint is loaded