The lightweight PyTorch wrapper for high-performance AI research. Scale your models, not the boilerplate.
*Codecov is > 90%+ but build delays may show less
PyTorch Lightning is just organized PyTorch
Lightning Design Philosophy
Lightning structures PyTorch code with these principles:
Lightning forces the following structure to your code which makes it reusable and shareable:
- Research code (the LightningModule).
- Engineering code (you delete, and is handled by the Trainer).
- Non-essential research code (logging, etc... this goes in Callbacks).
- Data (use PyTorch DataLoaders or organize them into a LightningDataModule).
Once you do this, you can train on multiple-GPUs, TPUs, CPUs, IPUs, HPUs and even in 16-bit precision without changing your code!
Get started with our 2 step guide
Lightning is rigorously tested across multiple CPUs, GPUs, TPUs, IPUs, and HPUs and against major Python and PyTorch versions.
Current build statuses
- ** tests run on two NVIDIA P100
- *** tests run on Google GKE TPUv2/3. TPU py3.7 means we support Colab and Kaggle env.
How To Use
Step 0: Install
Simple installation from PyPI
pip install pytorch-lightning
Other installation options
Install with optional dependencies
pip install pytorch-lightning['extra']
conda install pytorch-lightning -c conda-forge
Install stable 1.5.x
the actual status of 1.5 [stable] is following:
Install future release from the source
pip install git+https://github.com/PytorchLightning/pytorch-lightning.git@release/1.5.x --upgrade
Install bleeding-edge - future 1.6
Install nightly from the source (no guarantees)
pip install https://github.com/PyTorchLightning/pytorch-lightning/archive/master.zip
or from testing PyPI
pip install -iU https://test.pypi.org/simple/ pytorch-lightning
Step 1: Add these imports
import os import torch from torch import nn import torch.nn.functional as F from torchvision.datasets import MNIST from torch.utils.data import DataLoader, random_split from torchvision import transforms import pytorch_lightning as pl
Step 2: Define a LightningModule (nn.Module subclass)
A LightningModule defines a full system (ie: a GAN, autoencoder, BERT or a simple Image Classifier).
class LitAutoEncoder(pl.LightningModule): def __init__(self): super().__init__() self.encoder = nn.Sequential(nn.Linear(28 * 28, 128), nn.ReLU(), nn.Linear(128, 3)) self.decoder = nn.Sequential(nn.Linear(3, 128), nn.ReLU(), nn.Linear(128, 28 * 28)) def forward(self, x): # in lightning, forward defines the prediction/inference actions embedding = self.encoder(x) return embedding def training_step(self, batch, batch_idx): # training_step defines the train loop. It is independent of forward x, y = batch x = x.view(x.size(0), -1) z = self.encoder(x) x_hat = self.decoder(z) loss = F.mse_loss(x_hat, x) self.log("train_loss", loss) return loss def configure_optimizers(self): optimizer = torch.optim.Adam(self.parameters(), lr=1e-3) return optimizer
Note: Training_step defines the training loop. Forward defines how the LightningModule behaves during inference/prediction.
Step 3: Train!
dataset = MNIST(os.getcwd(), download=True, transform=transforms.ToTensor()) train, val = random_split(dataset, [55000, 5000]) autoencoder = LitAutoEncoder() trainer = pl.Trainer() trainer.fit(autoencoder, DataLoader(train), DataLoader(val))
Lightning has over 40+ advanced features designed for professional AI research at scale.
Here are some examples:
Highlighted feature code snippets
# 8 GPUs # no code changes needed trainer = Trainer(max_epochs=1, accelerator="gpu", devices=8) # 256 GPUs trainer = Trainer(max_epochs=1, accelerator="gpu", devices=8, num_nodes=32)
Train on TPUs without code changes
# no code changes needed trainer = Trainer(accelerator="tpu", devices=8)
# no code changes needed trainer = Trainer(precision=16)
from pytorch_lightning import loggers # tensorboard trainer = Trainer(logger=TensorBoardLogger("logs/")) # weights and biases trainer = Trainer(logger=loggers.WandbLogger()) # comet trainer = Trainer(logger=loggers.CometLogger()) # mlflow trainer = Trainer(logger=loggers.MLFlowLogger()) # neptune trainer = Trainer(logger=loggers.NeptuneLogger()) # ... and dozens more
es = EarlyStopping(monitor="val_loss") trainer = Trainer(callbacks=[es])
checkpointing = ModelCheckpoint(monitor="val_loss") trainer = Trainer(callbacks=[checkpointing])
Export to torchscript (JIT) (production use)
# torchscript autoencoder = LitAutoEncoder() torch.jit.save(autoencoder.to_torchscript(), "model.pt")
Export to ONNX (production use)
# onnx with tempfile.NamedTemporaryFile(suffix=".onnx", delete=False) as tmpfile: autoencoder = LitAutoEncoder() input_sample = torch.randn((1, 64)) autoencoder.to_onnx(tmpfile.name, input_sample, export_params=True) os.path.isfile(tmpfile.name)
Pro-level control of training loops (advanced users)
For complex/professional level work, you have optional full control of the training loop and optimizers.
class LitAutoEncoder(pl.LightningModule): def __init__(self): super().__init__() self.automatic_optimization = False def training_step(self, batch, batch_idx): # access your optimizers with use_pl_optimizer=False. Default is True opt_a, opt_b = self.optimizers(use_pl_optimizer=True) loss_a = ... self.manual_backward(loss_a, opt_a) opt_a.step() opt_a.zero_grad() loss_b = ... self.manual_backward(loss_b, opt_b, retain_graph=True) self.manual_backward(loss_b, opt_b) opt_b.step() opt_b.zero_grad()
Advantages over unstructured PyTorch
- Models become hardware agnostic
- Code is clear to read because engineering code is abstracted away
- Easier to reproduce
- Make fewer mistakes because lightning handles the tricky engineering
- Keeps all the flexibility (LightningModules are still PyTorch modules), but removes a ton of boilerplate
- Lightning has dozens of integrations with popular machine learning tools.
- Tested rigorously with every new PR. We test every combination of PyTorch and Python supported versions, every OS, multi GPUs and even TPUs.
- Minimal running speed overhead (about 300 ms per epoch compared with pure PyTorch).
In the Lightning 1.5 release, LightningLite now enables you to leverage all the capabilities of PyTorch Lightning Accelerators without any refactoring to your training loop. Check out the blogpost and docs for more info.
The lightning community is maintained by
- 10+ core contributors who are all a mix of professional engineers, Research Scientists, and Ph.D. students from top AI labs.
- 590+ active community contributors.
Want to help us build Lightning and reduce boilerplate for thousands of researchers? Learn how to make your first contribution here
Lightning is also part of the PyTorch ecosystem which requires projects to have solid testing, documentation and support.
Asking for help
If you have any questions please:
We're venture funded to make sure we can provide around the clock support, hire a full-time staff, attend conferences, and move faster through implementing features you request.
Grid AI is our platform for training models at scale on the cloud!
Sign up for our FREE community Tier here
To use grid, take your regular command:
python my_model.py --learning_rate 1e-6 --layers 2 --accelerator 'gpu' --devices 4
And change it to use the grid train command:
grid train --grid_gpus 4 my_model.py --learning_rate 'uniform(1e-6, 1e-1, 20)' --layers '[2, 4, 8, 16]'
The above command will launch (20 * 4) experiments each running on 4 GPUs (320 GPUs!) - by making ZERO changes to your code.