#1515th of 15 in ML frameworks and libraries
pytorch-lightning
PyTorch training framework that removes loop, device and distributed boilerplate
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 31.4k
- License
- Apache-2.0
- Last commit
- Sep 2026
- Last release
- Sep 2026
- Language
- Python
30,788 on 24 Mar 202631,406 on 13 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
PyTorch Lightning wraps PyTorch models in a LightningModule and a Trainer that handles the training loop, mixed precision, checkpointing, logging and multi-device scaling. The same package ships Lightning Fabric, a lower-level API that keeps your own loop and adds device placement and distributed strategies (DDP, FSDP, DeepSpeed). Runs on CPU, CUDA, Apple Silicon (MPS) and TPU.
Who it is for: ML engineers training or finetuning PyTorch models across GPUs or nodes
Strengths
- Same code scales from CPU to multi-node GPU or TPU by changing Trainer arguments
- Fabric offers a minimal-diff path for existing PyTorch loops, with DDP, FSDP and DeepSpeed
- Built-in early stopping and checkpointing callbacks; loggers for TensorBoard, W&B, Comet, MLflow
- Exports to TorchScript and ONNX; README cites about 300 ms per epoch overhead
Weaknesses
- Training only; serving needs the separate LitServe project
- Trainer requires restructuring code into LightningModule hooks
- README promotes Lightning Cloud, and most examples run as Lightning Studios
- Abstraction layer over PyTorch can complicate debugging unusual loops
What it needs
- GPU optional
- Needs PyTorch
- Models: PyTorch nn.Module, TorchScript export, ONNX export
Also in ML frameworks and libraries
See all 15| Rank | Project | Score |
|---|---|---|
| 1 | transformersModel-definition library for text, vision, audio and multimodal models | 94 out of 100 |
| 2 | pytorchPython tensor library with GPU support and tape-based autograd for deep learning | 92 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 86 out of 100 |
| #4 | pytorch-image-modelsPyTorch library of image models, pretrained weights and reference training scripts | 82 out of 100 |
| #5 | tensorflowTensor and neural network framework with Python and C++ APIs | 80 out of 100 |