def load(self, epoch_num):
load_path = self.get_path(epoch_num)
self.language_model = torch_utils.load_model(load_path).cuda()
self.init_trainer() # Need to reset optimizer
config['beam_size'] = 5
config['log_path'] = 'logs.txt'
config['save_directory'] = 'logs/squad_saved_data'
config['use_pretrained_embeddings'] = True
config['pretrained_embeddings_path'] = 'datasets/squad/word_embeddings.npy'
config['finetune_embeddings'] = False
config['load_model'] = False
config['gpu_mode'] = True
config[
'load_path'] = 'logs/squad_saved_data/model_6.pyt7' # CHANGE THIS TO WHATEVER PATH YOU WANT
io_utils.check_dir('logs/squad_saved_data')
language_model = LanguageModel(config)
if config['load_model']:
language_model = torch_utils.load_model(config['load_path'])
language_model.cuda()
language_wrapper = LanguageWrapper(language_model,
language_model_loader.get_vocab())
language_trainer = LanguageTrainer(config, language_wrapper,
language_model_loader)
for i in range(0, 15):
loss, accuracy, predictions = language_trainer.train(epoch_num=i)
if i % 3 == 2:
predictions = language_trainer.predict(
dataset_type=constants.DATASET_TEST,
epoch_num=10,
max_length=20,
from data_loaders.language_model_loader_truncate import LanguageModelLoaderTruncate
from models.language_model import LanguageModel
from models.language_trainer import LanguageTrainer
from models.language_wrapper import LanguageWrapper
from helpers import constants
import torch
from helpers import torch_utils, utils
from torch.autograd import variable
dataset_path = 'datasets/newsqa'
load_path = 'logs/squad_saved_data_truncated/model_2.pyt7'
language_model_loader = LanguageModelLoaderTruncate(
dataset_path, tokenizer_type=constants.TOKENIZER_TAB)
language_model = torch_utils.load_model(load_path).cuda()
language_model.config['save_directory'] = 'logs/newsqa_saved_data'
language_wrapper = LanguageWrapper(language_model,
language_model_loader.get_vocab())
language_trainer = LanguageTrainer(language_model.config, language_wrapper,
language_model_loader)
#test_predictions = language_trainer.predict(dataset_type=constants.DATASET_TEST,
# epoch_num=10, max_length=20)
#dev_predictions = language_trainer.predict(dataset_type=constants.DATASET_VALIDATION,
# epoch_num=10, max_length=10)
train_predictions = language_trainer.predict(
dataset_type=constants.DATASET_TRAIN, epoch_num=10, max_length=10)
utils.save_lines(
train_predictions,
from models.language_model import LanguageModel
from helpers import torch_utils
config = {}
config['vocab_size'] = 12
config['embedding_size'] = 20
config['hidden_size'] = 50
config['num_layers'] = 1
config['dropout'] = 0.0
config['batch_first'] = True
model = LanguageModel(config)
torch_utils.save_model(model, path='test.model')
model = torch_utils.load_model(path='test.model')