NVIDIA-Merlin/dataloader
GPU-optimized dataloader for training deep learning recommender models with TensorFlow, PyTorch, and JAX frameworks
Not currently ranked — collecting fresh signals.
star history
NVIDIA Merlin Dataloader accelerates training of recommender models by loading data directly into GPU memory and providing zero-copy transfers to deep learning frameworks via dlpack. It claims over 10x speedup over native framework dataloaders, handles datasets larger than memory, provides per-epoch shuffling, and supports distributed training across TensorFlow, PyTorch, and JAX.
Frequently asked
- What is NVIDIA-Merlin/dataloader?
- GPU-optimized dataloader for training deep learning recommender models with TensorFlow, PyTorch, and JAX frameworks
- Is dataloader open source?
- Yes — NVIDIA-Merlin/dataloader is open source, released under the Apache-2.0 license.
- What language is dataloader written in?
- NVIDIA-Merlin/dataloader is primarily written in Python.
- How popular is dataloader?
- NVIDIA-Merlin/dataloader has 423 stars on GitHub.
- Where can I find dataloader?
- NVIDIA-Merlin/dataloader is on GitHub at https://github.com/NVIDIA-Merlin/dataloader.