def create_model(model_class, dicts, gen_out_size):
if opt.data_type == 'code':
encoder = lib.TreeEncoder(opt, dicts["src"])
decoder = lib.TreeDecoder(opt, dicts["tgt"])
elif opt.data_type == 'text':
encoder = lib.Encoder(opt, dicts["src"])
decoder = lib.TreeDecoder(opt, dicts["tgt"])
elif opt.data_type == 'hybrid':
code_encoder = lib.TreeEncoder(opt, dicts["src"])
text_encoder = lib.Encoder(opt, dicts["src"])
decoder = lib.HybridDecoder(opt, dicts["tgt"])
# Use memory efficient generator when output size is large and
# max_generator_batches is smaller than batch_size.
if opt.max_generator_batches < opt.batch_size and gen_out_size > 1:
generator = lib.MemEfficientGenerator(nn.Linear(opt.rnn_size, gen_out_size), opt)
else:
generator = lib.BaseGenerator(nn.Linear(opt.rnn_size, gen_out_size), opt)
if opt.data_type == 'code' or opt.data_type == 'text':
model = model_class(encoder, decoder, generator, opt)
elif opt.data_type == 'hybrid':
model = model_class(code_encoder, text_encoder, decoder, generator, opt)
init(model)
optim = create_optim(model)
return model, optim
def create_model(model_class, dicts, gen_out_size):
encoder = lib.Encoder(opt, dicts["src"])
decoder = lib.Decoder(opt, dicts["tgt"])
generator = lib.BaseGenerator(nn.Linear(opt.rnn_size, gen_out_size), opt)
model = model_class(encoder, decoder, generator, opt)
init(model)
optim = create_optim(model)
return model, optim
def create_model(model_class, dicts, gen_out_size):
encoder = lib.Encoder(opt, dicts["src"])
decoder = lib.Decoder(opt, dicts["tgt"])
# Use memory efficient generator when output size is large and
# max_generator_batches is smaller than batch_size.
if opt.max_generator_batches < opt.batch_size and gen_out_size > 1:
generator = lib.MemEfficientGenerator(nn.Linear(opt.rnn_size, gen_out_size), opt)
else:
generator = lib.BaseGenerator(nn.Linear(opt.rnn_size, gen_out_size), opt)
model = model_class(encoder, decoder, generator, opt)
init(model)
optim = create_optim(model)
return model, optim