| import datasets |
| import pytorch_lightning as pl |
|
|
| from s1.QuantizeDataset import QuantizeDataset |
| from s1.sampler import RandomBucketSampler |
| from torch.utils import data |
|
|
|
|
| class QuantizeDataModule(pl.LightningDataModule): |
| def __init__(self, hp, data_dir='dump/phone_semantic_dataset'): |
| super().__init__() |
| self.hp = hp |
| dataset_all = datasets.load_from_disk(data_dir, keep_in_memory=True) |
| self.train_data = QuantizeDataset(hp, dataset_all['train']) |
| self.val_data = QuantizeDataset(hp, dataset_all['test']) |
|
|
|
|
| def train_dataloader(self): |
| length = self.train_data.lengths |
| sampler = RandomBucketSampler(self.hp.train_bucket_size, length, self.hp.batch_size, drop_last=True, distributed=self.hp.distributed, |
| world_size=self.trainer.world_size, rank=self.trainer.local_rank) |
| dataset = data.DataLoader(self.train_data, |
| num_workers=self.hp.nworkers, |
| batch_sampler=sampler, |
| collate_fn=self.train_data.seqCollate) |
| return dataset |
|
|
| def val_dataloader(self): |
| dataset = data.DataLoader(self.val_data, |
| num_workers=self.hp.nworkers, |
| collate_fn=self.val_data.seqCollate, |
| shuffle=False) |
| return dataset |
