def __init__(self, args):
self.args = args
self.batch_size = args.batch_size
#self.batch_size = 1
self.finetune_topk = args.finetune_topk
self.lr = args.lr
self.use_cuda = (args.use_cuda == True) and torch.cuda.is_available()
print('Use cuda:', self.use_cuda)
if self.use_cuda:
torch.cuda.set_device(int(args.gpu))
module = importlib.import_module(".".join([args.model_name]))
MyModel = getattr(module, 'MyModel')
self.network = MyModel(args)
self.init_optimizer()
if args.pretrained:
print('Load pretrained model from %s...' % args.pretrained)
self.load(args.pretrained)
else:
self.load_embeddings(vocab.tokens(), args.embedding_file)
self.network.register_buffer(
'fixed_embedding',
self.network.embedding.weight.data[self.finetune_topk:].clone())
if self.use_cuda:
self.network.cuda()
print(self.network)
self._report_num_trainable_parameters()
def __init__(self, args):
self.args = args
self.batch_size = args.batch_size
self.finetune_topk = args.finetune_topk
self.lr = args.lr
self.use_cuda = (args.use_cuda == True) and torch.cuda.is_available()
print('Use cuda:', self.use_cuda)
if self.use_cuda:
torch.cuda.set_device(int(args.gpu))
self.network = TriAN(args)
self.init_optimizer()
# load pretrained model
if args.pretrained:
print('Load pretrained model from %s...' % args.pretrained)
self.load(args.pretrained)
else:
if args.use_elmo == False:
self.load_embeddings(vocab.tokens(), args.embedding_file)
if args.use_elmo == False:
self.network.register_buffer(
'fixed_embedding',
self.network.embedding.weight.data[self.finetune_topk:].clone(
))
self.elmo = None
else:
options_file = "https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_options.json"
weight_file = "https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5"
self.elmo = ElmoEmbedder(options_file, weight_file, cuda_device=0)
# self.elmo = Elmo(options_file, weight_file, num_output_representations=2, dropout=0, requires_grad=False)
if self.use_cuda:
self.network.cuda()
print(self.network)
self._report_num_trainable_parameters()
def __init__(self, args):
self.args = args
self.batch_size = args.batch_size
self.finetune_topk = args.finetune_topk
self.lr = args.lr
#self.use_cuda = (args.use_cuda == True) and torch.cuda.is_available()
#print('Use cuda:', self.use_cuda)
#if self.use_cuda:
# torch.cuda.set_device(int(args.gpu))
tf.reset_default_graph()
self.sess = tf.Session()
self.network = TriAN(args)
self.init_optimizer()
if args.pretrained:
print('Load pretrained model from %s...' % args.pretrained)
self.load(args.pretrained)
else:
self.word_emb_mat = self.load_embeddings(vocab.tokens(),
args.embedding_file)
self.sess.run(tf.global_variables_initializer())
self._report_num_trainable_parameters()
def __init__(self, args):
self.args = args
self.batch_size = args.batch_size
self.finetune_topk = args.finetune_topk
self.lr = args.lr
self.use_cuda = (args.use_cuda == True) and torch.cuda.is_available()
print('Use cuda:', self.use_cuda)
if self.use_cuda:
torch.cuda.set_device(int(args.gpu))
self.network = JointQA(args)
self.init_optimizer()
self.nli_loss_criterion = torch.nn.NLLLoss()
if args.pretrained:
print('Load pretrained model from %s...' % args.pretrained)
self.load(args.pretrained)
else:
self.load_embeddings(vocab.tokens(), args.embedding_file)
self.network.register_buffer('fixed_embedding', self.network.embedding.weight.data[self.finetune_topk:].clone())
if self.use_cuda:
self.network.cuda()
print(self.network)
self._report_num_trainable_parameters()
def __init__(self, args):
self.args = args
self.batch_size = args.batch_size
self.finetune_topk = args.finetune_topk
self.lr = args.lr
self.use_cuda = (args.use_cuda == True) and torch.cuda.is_available()
print('Use cuda:', self.use_cuda)
if self.use_cuda:
torch.cuda.set_device(int(args.gpu))
# self.network = TriAN(args)
# self.network = simpleModel(args)
self.network = simpleModel(args)
self.init_optimizer()
# load pretrained model
if args.pretrained:
print('Load pretrained model from %s...' % args.pretrained)
self.load(args.pretrained)
else:
if args.use_elmo == False:
self.load_embeddings(vocab.tokens(), args.embedding_file)
if args.use_elmo == False:
self.network.register_buffer('fixed_embedding',
self.network.embedding.weight.data[self.finetune_topk:].clone())
self.elmo = None
else:
options_file = "https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_options.json"
weight_file = "https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5"
from allennlp.commands.elmo import ElmoEmbedder
self.elmo = ElmoEmbedder(options_file, weight_file, cuda_device=0)
if self.use_cuda:
self.network.cuda()
print(self.network)
self._report_num_trainable_parameters()
self.crit = torch.nn.BCEWithLogitsLoss(reduce=False, pos_weight=torch.FloatTensor([args.pos_weight]).cuda())