Lightning in 3 steps¶
In this guide we’ll show you how to organize your PyTorch code into Lightning in 3 simple steps.
Organizing your code with PyTorch Lightning makes your code:
Keep all the flexibility (this is all pure PyTorch), but removes a ton of boilerplate
More readable by decoupling the research code from the engineering
Easier to reproduce
Less error prone by automating most of the training loop and tricky engineering
Scalable to any hardware without changing your model
Here’s a 2 minute conversion guide for PyTorch projects:
Step 0: Install PyTorch Lightning¶
You can install using pip
pip install pytorch-lightning
Or with conda (see how to install conda here):
conda install pytorch-lightning -c conda-forge
You could also use conda environments
conda activate my_env
pip install pytorch-lightning
Step 1: Define LightningModule¶
import os
import torch
import torch.nn.functional as F
from torchvision.datasets import MNIST
from torchvision import transforms
from torch.utils.data import DataLoader
import pytorch_lightning as pl
from torch.utils.data import random_split
class LitModel(pl.LightningModule):
def __init__(self):
super().__init__()
self.layer_1 = torch.nn.Linear(28 * 28, 128)
self.layer_2 = torch.nn.Linear(128, 10)
def forward(self, x):
x = x.view(x.size(0), -1)
x = self.layer_1(x)
x = F.relu(x)
x = self.layer_2(x)
return x
def configure_optimizers(self):
optimizer = torch.optim.Adam(self.parameters(), lr=1e-3)
return optimizer
def training_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
result = pl.TrainResult(loss)
return result
def validation_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
result = pl.EvalResult(checkpoint_on=loss)
result.log('val_loss', loss)
return result
def test_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
result = pl.EvalResult()
result.log('test_loss', loss)
return result
The LightningModule holds your research code:
The Train loop
The Validation loop
The Test loop
The Model + system architecture
The Optimizer
A LightningModule is a torch.nn.Module but with added functionality.
It organizes your research code into Model Hooks.
In the snippet above we override the basic hooks, but a full list of hooks to customize can be found under Model Hooks.
You can use your LightningModule just like a PyTorch model.
model = LitModel()
model.eval()
y_hat = model(x)
model.anything_you_can_do_with_pytorch()
More details in LightningModule docs.
Convert your PyTorch Module to Lightning¶
1. Move your computational code¶
Move the model architucture and forward pass to your LightningModule.
class LitModel(pl.LightningModule):
def __init__(self):
super().__init__()
self.layer_1 = torch.nn.Linear(28 * 28, 128)
self.layer_2 = torch.nn.Linear(128, 10)
def forward(self, x):
x = x.view(x.size(0), -1)
x = self.layer_1(x)
x = F.relu(x)
x = self.layer_2(x)
return x
2. Move the optimizer(s) and schedulers¶
Move your optimizers to pytorch_lightning.core.LightningModule.configure_optimizers() hook. Make sure to use the hook parameters (self in this case).
class LitModel(pl.LightningModule):
def configure_optimizers(self):
optimizer = torch.optim.Adam(self.parameters(), lr=1e-3)
return optimizer
3. Find the train loop “meat”¶
Lightning automates most of the trining for you, the epoch and batch iterations, all you need to keep is the training step logic. This should go into pytorch_lightning.core.LightningModule.training_step() hook (make sure to use the hook parameters, self in this case):
class LitModel(pl.LightningModule):
def training_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
return loss
4. Find the val loop “meat”¶
Lightning automates the validation (enabling gradients in the train loop and disabling in eval). To add an (optional) validation loop add logic to pytorch_lightning.core.LightningModule.validation_step() hook (make sure to use the hook parameters, self in this case):
class LitModel(LightningModule):
def validation_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
val_loss = F.cross_entropy(y_hat, y)
return val_loss
5. Find the test loop “meat”¶
You might also need an optional test loop. Add the following callback to your LightningModule
class LitModel(pl.LightningModule):
def test_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
result = pl.EvalResult()
result.log('test_loss', loss)
return result
Note
The test loop is not automated in Lightning. You will need to specifically call test (this is done so you don’t use the test set by mistake).
6. Remove any .cuda() or to.device() calls¶
Your LightningModule can automatically run on any hardware!
7. Wrap loss in a TrainResult/EvalResult¶
Instead of returning the loss you can also use TrainResult and EvalResult, plain Dict objects that give you options for logging on every step and/or at the end of the epoch.
It also allows logging to the progress bar (by setting prog_bar=True). Read more in Result.
class LitModel(pl.LightningModule):
def training_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
result = pl.TrainResult(loss)
# Add logging to progress bar (note that efreshing the progress bar too frequently
# in Jupyter notebooks or Colab may freeze your UI)
result.log('train_loss', loss, prog_bar=True)
return result
def validation_step(self, batch, batch_idx):
x, y = batch
y_hat = self(x)
loss = F.cross_entropy(y_hat, y)
# Checkpoint model based on validation loss
result = pl.EvalResult(checkpoint_on=loss)
result.log('val_loss', loss)
return result
8. Override default callbacks¶
A LightningModule handles advances cases by allowing you to override any critical part of training
via Model Hooks that are called on your LightningModule.
class LitModel(pl.LightningModule):
def backward(self, trainer, loss, optimizer, optimizer_idx):
loss.backward()
def optimizer_step(self, epoch, batch_idx,
optimizer, optimizer_idx,
second_order_closure,
on_tpu, using_native_amp, using_lbfgs):
optimizer.step()
For certain train/val/test loops, you may wish to do more than just logging. In this case, you can also implement __epoch_end which gives you the output for each step
Here’s the motivating Pytorch example:
validation_step_outputs = []
for batch_idx, batch in val_dataloader():
out = validation_step(batch, batch_idx)
validation_step_outputs.append(out)
validation_epoch_end(validation_step_outputs)
And the lightning equivalent
class LitModel(pl.LightningModule):
def validation_step(self, batch, batch_idx):
loss = ...
predictions = ...
result = pl.EvalResult(checkpoint_on=loss)
result.log('val_loss', loss)
result.predictions = predictions
def validation_epoch_end(self, validation_step_outputs):
all_val_losses = validation_step_outputs.val_loss
all_predictions = validation_step_outputs.predictions
Step 2: Fit with Lightning Trainer¶
# dataloaders
dataset = MNIST(os.getcwd(), download=True, transform=transforms.ToTensor())
train, val = random_split(dataset, [55000, 5000])
train_loader = DataLoader(train)
val_loader = DataLoader(val)
# init model
model = LitModel()
# most basic trainer, uses good defaults (auto-tensorboard, checkpoints, logs, and more)
trainer = pl.Trainer()
trainer.fit(model, train_loader, val_loader)
Init LightningModule, your PyTorch dataloaders, and then the PyTorch Lightning Trainer.
The Trainer will automate:
The epoch iteration
The batch iteration
The calling of optimizer.step()
Logging to Tensorboard (see Loggers options)
Multi-GPU training support
16-bit training support
All automated code is rigorously tested and benchmarked.
Check out more flags in the Trainer docs.
Using CPUs/GPUs/TPUs¶
It’s trivial to use CPUs, GPUs or TPUs in Lightning. There’s NO NEED to change your code, simply change the Trainer options.
# train on 1024 CPUs across 128 machines
trainer = pl.Trainer(
num_processes=8,
num_nodes=128
)
# train on 1 GPU
trainer = pl.Trainer(gpus=1)
# train on 256 GPUs
trainer = pl.Trainer(
gpus=8,
num_nodes=32
)
# Multi GPU with mixed precision
trainer = pl.Trainer(gpus=2, precision=16)
# Train on TPUs
trainer = pl.Trainer(tpu_cores=8)
Without changing a SINGLE line of your code, you can now do the following with the above code:
# train on TPUs using 16 bit precision with early stopping
# using only half the training data and checking validation every quarter of a training epoch
trainer = pl.Trainer(
tpu_cores=8,
precision=16,
early_stop_callback=True,
limit_train_batches=0.5,
val_check_interval=0.25
)
Step 3: Define Your Data¶
Lightning works with pure PyTorch DataLoaders
train_dataloader = DataLoader(...)
val_dataloader = DataLoader(...)
trainer.fit(model, train_dataloader, val_dataloader)
Optional: DataModule¶
DataLoader and data processing code tends to end up scattered around.
Make your data code more reusable by organizing
it into a LightningDataModule
class MNISTDataModule(pl.LightningDataModule):
def __init__(self, batch_size=32):
super().__init__()
self.batch_size = batch_size
# When doing distributed training, Datamodules have two optional arguments for
# granular control over download/prepare/splitting data:
# OPTIONAL, called only on 1 GPU/machine
def prepare_data(self):
MNIST(os.getcwd(), train=True, download=True)
MNIST(os.getcwd(), train=False, download=True)
# OPTIONAL, called for every GPU/machine (assigning state is OK)
def setup(self, stage):
# transforms
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])
# split dataset
if stage == 'fit':
mnist_train = MNIST(os.getcwd(), train=True, transform=transform)
self.mnist_train, self.mnist_val = random_split(mnist_train, [55000, 5000])
if stage == 'test':
mnist_test = MNIST(os.getcwd(), train=False, transform=transform)
self.mnist_test = MNIST(os.getcwd(), train=False, download=True)
# return the dataloader for each split
def train_dataloader(self):
mnist_train = DataLoader(self.mnist_train, batch_size=self.batch_size)
return mnist_train
def val_dataloader(self):
mnist_val = DataLoader(self.mnist_val, batch_size=self.batch_size)
return mnist_val
def test_dataloader(self):
mnist_test = DataLoader(mnist_test, batch_size=self.batch_size)
return mnist_test
LightningDataModule is designed to enable sharing and reusing data splits
and transforms across different projects. It encapsulates all the steps needed to process data: downloading,
tokenizeing, processing etc.
Now you can simply pass your LightningDataModule to
the Trainer:
# init model
model = LitModel()
# init data
dm = MNISTDataModule()
# train
trainer = pl.Trainer()
trainer.fit(model, dm)
# test
trainer.test(datamodule=dm)
DataModules are specifically useful for building models based on data. Read more on LightningDataModule.
Learn more¶
That’s it! Once you build your module, data, and call trainer.fit(), Lightning trainer calls each loop at the correct time as needed.
You can then boot up your logger or tensorboard instance to view training logs
tensorboard --logdir ./lightning_logs
Advanced Lightning Features¶
Once you define and train your first Lightning model, you might want to try other cool features like
Add custom callbacks (self-contained programs that can be reused across projects)
Dry run mode (Hit every line of your code once to see if you have bugs, instead of waiting hours to crash on validation ;)
Use multiple optimizers to do Reinforcement learning or even GANs
Or read our Step-by-step walk-through to learn more!
