def test():
saved_model_path = './model.pth' # lightning_logs/version_N/checkpoints/* 最优模型软链接
model = BertClassifier.load_from_checkpoint(saved_model_path)
model.eval()
print(model)
trainer = Trainer(gpus=1)
result = trainer.test(model)
print(result)
model_path = str(list(exp_dir.glob("**/*.ckpt"))[0])
hparams_path = str(exp_dir / "hparams.yaml")
# load model
model = Distiller.load_from_checkpoint(
model_path,
hparams_file=hparams_path,
map_location=device,
)
model.eval()
model.freeze()
return model
def load_data(data_dir, num_workers, hparams):
hparams.dataset.path = data_dir
hparams.dataset.on_memory = True
hparams.dataset.num_workers = num_workers
return TripleEmbeddingDataModule(hparams)
if __name__ == "__main__":
model = load_model(args.exp_dir)
dm = load_data(args.data_dir, args.num_workers, model.hparams)
# dm.prepare_data()
dm.setup("test")
# trainer for test
trainer = Trainer()
trainer.test(model, datamodule=dm)