>>> from datasets import Dataset
>>> data = [[1, 2],[3, 4]]
>>> ds = Dataset.from_dict({"data": data})
>>> ds = ds.with_format("torch")
>>> ds[0]
{'data': tensor([1, 2])}
>>> ds[:2]
{'data': tensor([[1, 2],
         [3, 4]])}

>>> import torch
>>> device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
>>> ds = ds.with_format("torch", device=device)
>>> ds[0]
{'data': tensor([1, 2], device='cuda:0')}

>>> from datasets import Dataset
>>> data = [[[1, 2],[3, 4]],[[5, 6],[7, 8]]]  # fixed shape
>>> ds = Dataset.from_dict({"data": data})
>>> ds = ds.with_format("torch")
>>> ds[0]
{'data': tensor([[1, 2],
         [3, 4]])}

>>> from datasets import Dataset
>>> data = [[[1, 2],[3]],[[4, 5, 6],[7, 8]]]  # varying shape
>>> ds = Dataset.from_dict({"data": data})
>>> ds = ds.with_format("torch")
>>> ds[0]
{'data': [tensor([1, 2]), tensor([3])]}

>>> from datasets import Dataset, Features, Array2D
>>> data = [[[1, 2],[3, 4]],[[5, 6],[7, 8]]]
>>> features = Features({"data": Array2D(shape=(2, 2), dtype='int32')})
>>> ds = Dataset.from_dict({"data": data}, features=features)
>>> ds = ds.with_format("torch")
>>> ds[0]
{'data': tensor([[1, 2],
         [3, 4]])}
>>> ds[:2]
{'data': tensor([[[1, 2],
          [3, 4]],
 
         [[5, 6],
          [7, 8]]])}

>>> from datasets import Dataset, Features, ClassLabel
>>> labels = [0, 0, 1]
>>> features = Features({"label": ClassLabel(names=["negative", "positive"])})
>>> ds = Dataset.from_dict({"label": labels}, features=features) 
>>> ds = ds.with_format("torch")  
>>> ds[:3]
{'label': tensor([0, 0, 1])}

>>> from datasets import Dataset, Features, Audio, Image
>>> images = ["path/to/image.png"] * 10
>>> features = Features({"image": Image()})
>>> ds = Dataset.from_dict({"image": images}, features=features) 
>>> ds = ds.with_format("torch")
>>> ds[0]["image"].shape
torch.Size([512, 512, 4])
>>> ds[0]
{'image': tensor([[[255, 215, 106, 255],
         [255, 215, 106, 255],
         ...,
         [255, 255, 255, 255],
         [255, 255, 255, 255]]], dtype=torch.uint8)}
>>> ds[:2]["image"].shape
torch.Size([2, 512, 512, 4])
>>> ds[:2]
{'image': tensor([[[[255, 215, 106, 255],
          [255, 215, 106, 255],
          ...,
          [255, 255, 255, 255],
          [255, 255, 255, 255]]]], dtype=torch.uint8)}

>>> from datasets import Dataset, Features, Audio, Image
>>> audio = ["path/to/audio.wav"] * 10
>>> features = Features({"audio": Audio()})
>>> ds = Dataset.from_dict({"audio": audio}, features=features) 
>>> ds = ds.with_format("torch")  
>>> ds[0]["audio"]["array"]
tensor([ 6.1035e-05,  1.5259e-05,  1.6785e-04,  ..., -1.5259e-05,
        -1.5259e-05,  1.5259e-05])
>>> ds[0]["audio"]["sampling_rate"]
tensor(44100)

>>> import numpy as np
>>> from datasets import Dataset 
>>> from torch.utils.data import DataLoader
>>> data = np.random.rand(16)
>>> label = np.random.randint(0, 2, size=16)
>>> ds = Dataset.from_dict({"data": data, "label": label}).with_format("torch")
>>> dataloader = DataLoader(ds, batch_size=4)
>>> for batch in dataloader:
...     print(batch)                                                                                            
{'data': tensor([0.0047, 0.4979, 0.6726, 0.8105]), 'label': tensor([0, 1, 0, 1])}
{'data': tensor([0.4832, 0.2723, 0.4259, 0.2224]), 'label': tensor([0, 0, 0, 0])}
{'data': tensor([0.5837, 0.3444, 0.4658, 0.6417]), 'label': tensor([0, 1, 0, 0])}
{'data': tensor([0.7022, 0.1225, 0.7228, 0.8259]), 'label': tensor([1, 1, 1, 1])}

>>> import numpy as np
>>> from datasets import Dataset, load_from_disk
>>> from torch.utils.data import DataLoader
>>> data = np.random.rand(10_000)
>>> Dataset.from_dict({"data": data}).save_to_disk("my_dataset")
>>> ds = load_from_disk("my_dataset").with_format("torch")
>>> dataloader = DataLoader(ds, batch_size=32, num_workers=4)

>>> import numpy as np
>>> from datasets import Dataset, load_dataset
>>> from torch.utils.data import DataLoader
>>> data = np.random.rand(10_000)
>>> Dataset.from_dict({"data": data}).push_to_hub("<username>/my_dataset")  # Upload to the Hugging Face Hub
>>> my_iterable_dataset = load_dataset("<username>/my_dataset", streaming=True, split="train")
>>> dataloader = DataLoader(my_iterable_dataset, batch_size=32)

>>> my_iterable_dataset = load_dataset("deepmind/code_contests", streaming=True, split="train")
>>> my_iterable_dataset.num_shards
39
>>> dataloader = DataLoader(my_iterable_dataset, batch_size=32, num_workers=4)

>>> from torchdata.stateful_dataloader import StatefulDataLoader
>>> my_iterable_dataset = load_dataset("deepmind/code_contests", streaming=True, split="train")
>>> dataloader = StatefulDataLoader(my_iterable_dataset, batch_size=32, num_workers=4)
>>> # save in the middle of training
>>> state_dict = dataloader.state_dict()
>>> # and resume later
>>> dataloader.load_state_dict(state_dict)

import os
from datasets.distributed import split_dataset_by_node

ds = split_dataset_by_node(ds, rank=int(os.environ["RANK"]), world_size=int(os.environ["WORLD_SIZE"]))

from torch.distributed.device_mesh import init_device_mesh
from datasets.distributed import split_dataset_by_node

# Hybrid sharding spreads data parallelism over two mesh dimensions, but both
# need different data per rank, so `dp_size` is their product.
mesh = init_device_mesh("cuda", (2, 2, 2), mesh_dim_names=("dp_replicate", "fsdp", "tp"))
dp_replicate_mesh, fsdp_mesh = mesh["dp_replicate"], mesh["fsdp"]

dp_size = dp_replicate_mesh.size() * fsdp_mesh.size()
dp_rank = dp_replicate_mesh.get_local_rank() * shard.size() + fsdp_mesh.get_local_rank()

ds = split_dataset_by_node(ds, rank=dp_rank, world_size=dp_size)