Jax Dataloader: Unified Loading for PyTorch, TF, and Hugging Face

BirkhoffG developed this utility to provide PyTorch-like dataloading for JAX, with the most recent update in December 2025. It enables the ingestion of data from 3 major backends—PyTorch, TensorFlow, and Hugging Face—into JAX-compatible iterators.

Use Cases

• Loading huggingface-datasets into JAX training loops
• Converting torch.utils.data.Dataset objects for JAX model ingestion
• Streaming tensorflow_datasets (TFDS) into JAX-based neural networks

Strengths

• Supports 3 major framework backends (PyTorch, TensorFlow, Hugging Face)
• Apache-2.0 license for open commercial use
• Implements PyTorch-style syntax for JAX users

Limitations

• Contains no internal data records, requiring external dataset dependencies
• Performance is limited by the data fetching efficiency of the underlying framework

Provenance

Source

GitHub (BirkhoffG)

Freshness

Last updated December 2025.