model = BayesianUniSkip('data/skip_thoughts', word_to_idx.keys())
for param in model.parameters():
param.requires_grad = False
elif args.use_bert:
model = BertModel.from_pretrained('bert-base-uncased')
model.eval()
elif args.use_gpt:
model = OpenAIGPTModel.from_pretrained('openai-gpt')
model.eval()
else:
model = RnnEncoder(dict_size=len(word_to_idx),
embed_size=args.embed_size,
hidden_dim=args.rnn_hidden_dim,
drop_prob=0.5)
generator = Generator()
discriminator = Discriminator()
dataloader = torch.utils.data.DataLoader(train_val_dataset,
batch_size=1,
shuffle=True)
model = model.to(device)
generator = generator.to(device)
discriminator = discriminator.to(device)
trainer = Trainer(dataloader, model, generator, discriminator, None, None, 1, device, None)
print("Loading model files")
trainer.load_model(args.model_path)
print("Generating image from text")
trainer.generate(args.text, args.output_file, args.count)
print("Image saved to {}".format(args.output_file))
