def _get_pretrained_model(self):
config = {
"checkpoint": self._get_checkpoint(),
"seed": 42,
"data_dir": "./data",
"glove_max_vectors": 10000,
"glove_dim": 300,
"batch_size": 64,
"cuda": True,
"encoder": self.encoder,
"max_epochs": None,
"lstm_hidden_dim": 2048 if self.encoder != "awe" else 300,
"classifier_hidden_dim": 512,
"debug": False
}
config = argparse.Namespace(**config)
pl.seed_everything(config.seed)
data_module = SNLIDataModule(config)
data_module.setup()
encoder = get_encoder(config)
model = InferSent.load_from_checkpoint(
config.checkpoint,
embeddings=data_module.glove_embeddings(),
encoder=encoder,
config=config)
model.to("cuda")
return model, data_module
parser.add_argument("--classifier_hidden_dim", type=int, default=512)
parser.add_argument("--debug",
action="store_true",
help="Only use 1%% of the training data.")
config = parser.parse_args()
pl.seed_everything(config.seed)
data_module = SNLIDataModule(config)
data_module.setup()
encoder = get_encoder(config)
model = InferSent.load_from_checkpoint(
config.checkpoint,
embeddings=data_module.glove_embeddings(),
encoder=encoder,
config=config)
model.to("cuda")
model.eval()
PATH_TO_SENTEVAL = '../SentEval'
PATH_TO_DATA = '../SentEval/data'
sys.path.insert(0, PATH_TO_SENTEVAL)
params_senteval = {
'task_path': PATH_TO_DATA,
'usepytorch': True,
'kfold': 2
}
params_senteval['classifier'] = {
