← All notes

Short answer

Datasets return Python objects by default. Call with_format("torch") (or "tf") to get framework tensors, optionally on a chosen device. [S1][S2]

Points

• A Dataset wraps an Arrow table. [S1]
• device can place tensors on a GPU. [S1]
• The TensorFlow guide shows with_format("tf") yielding tf.Tensor values. [S2]

Claims

• id: C1 claim: Datasets return regular Python objects by default. source: S1 evidence: "By default, datasets return regular python objects"
• id: C2 claim: with_format switches the dataset to pytorch tensors. source: S1 evidence: "set the format of the dataset to pytorch using Dataset.with_format()"
• id: C3 claim: A Dataset wraps an Arrow table. source: S1 evidence: "A Dataset object is a wrapper of an Arrow table"
• id: C4 claim: with_format tf yields TensorFlow tensors. source: S2 evidence: "with_format( "tf" )"

Sources

• [S1] Use with PyTorch · Hugging Face. Hugging Face (Datasets docs). https://huggingface.co/docs/datasets/use_with_pytorch . License: Apache-2.0 (huggingface/datasets docs). Accessed 2026-11-05.
• [S2] Use with TensorFlow. Hugging Face. https://huggingface.co/docs/datasets/use_with_tensorflow . License: Apache-2.0 (huggingface/datasets docs). Accessed 2026-11-05.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted).

Source notice: Hugging Face Datasets documentation, Apache License 2.0 (https://github.com/huggingface/datasets/blob/main/LICENSE).

@misc{iil_note_e3njqxdx,
  author = {{iinventedlight}},
  title = {{Use Hugging Face Datasets with PyTorch}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-04},
  url = {https://iinventedlight.com/library/notes/hf-datasets-pytorch/},
  note = {Permanent ID iil:note:e3njqxdx. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt. This page in Markdown: /library/notes/hf-datasets-pytorch.md.