Fabric is the fast and lightweight way to scale PyTorch models without boilerplate

Lightning Fabric: Expert control.

Run on any device at any scale with expert-level control over PyTorch training loop and scaling strategy. You can even write your own Trainer.

Fabric is designed for the most complex models like foundation model scaling, LLMs, diffusion, transformers, reinforcement learning, active learning. Of any size.

+ import lightning as L
  import torch; import torchvision as tv

  dataset = tv.datasets.CIFAR10("data", download=True,
                                train=True,
                                transform=tv.transforms.ToTensor())

+ fabric = L.Fabric()
+ fabric.launch()

  model = tv.models.resnet18()
  optimizer = torch.optim.SGD(model.parameters(), lr=0.001)
- device = "cuda" if torch.cuda.is_available() else "cpu"
- model.to(device)
+ model, optimizer = fabric.setup(model, optimizer)

  dataloader = torch.utils.data.DataLoader(dataset, batch_size=8)
+ dataloader = fabric.setup_dataloaders(dataloader)

  model.train()
  num_epochs = 10
  for epoch in range(num_epochs):
      for batch in dataloader:
          inputs, labels = batch
-         inputs, labels = inputs.to(device), labels.to(device)
          optimizer.zero_grad()
          outputs = model(inputs)
          loss = torch.nn.functional.cross_entropy(outputs, labels)
-         loss.backward()
+         fabric.backward(loss)
          optimizer.step()

import lightning as L
import torch; import torchvision as tv

dataset = tv.datasets.CIFAR10("data", download=True,
                              train=True,
                              transform=tv.transforms.ToTensor())

fabric = L.Fabric()
fabric.launch()

model = tv.models.resnet18()
optimizer = torch.optim.SGD(model.parameters(), lr=0.001)
model, optimizer = fabric.setup(model, optimizer)

dataloader = torch.utils.data.DataLoader(dataset, batch_size=8)
dataloader = fabric.setup_dataloaders(dataloader)

model.train()
num_epochs = 10
for epoch in range(num_epochs):
    for batch in dataloader:
        inputs, labels = batch
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = torch.nn.functional.cross_entropy(outputs, labels)
        fabric.backward(loss)
        optimizer.step()

Key features

# Use your available hardware
# no code changes needed
fabric = Fabric()

# Run on GPUs (CUDA or MPS)
fabric = Fabric(accelerator="gpu")

# 8 GPUs
fabric = Fabric(accelerator="gpu", devices=8)

# 256 GPUs, multi-node
fabric = Fabric(accelerator="gpu", devices=8, num_nodes=32)

# Run on TPUs
fabric = Fabric(accelerator="tpu")

# Use state-of-the-art distributed training techniques
fabric = Fabric(strategy="ddp")
fabric = Fabric(strategy="deepspeed")
fabric = Fabric(strategy="fsdp")

# Switch the precision
fabric = Fabric(precision="16-mixed")
fabric = Fabric(precision="64")

  # no more of this!
- model.to(device)
- batch.to(device)

import lightning as L


class MyCustomTrainer:
    def __init__(self, accelerator="auto", strategy="auto", devices="auto", precision="32-true"):
        self.fabric = L.Fabric(accelerator=accelerator, strategy=strategy, devices=devices, precision=precision)

    def fit(self, model, optimizer, dataloader, max_epochs):
        self.fabric.launch()

        model, optimizer = self.fabric.setup(model, optimizer)
        dataloader = self.fabric.setup_dataloaders(dataloader)
        model.train()

        for epoch in range(max_epochs):
            for batch in dataloader:
                input, target = batch
                optimizer.zero_grad()
                output = model(input)
                loss = loss_fn(output, target)
                self.fabric.backward(loss)
                optimizer.step()

You can find a more extensive example in our examples

Continuous Integration

Lightning is rigorously tested across multiple CPUs and GPUs and against major Python and PyTorch versions.

*Codecov is > 90%+ but build delays may show less

Getting started

Install Lightning

TIP: We strongly recommend creating a virtual environment first. Don’t know what this is? Follow our beginner guide here.

• Python 3.8.x or later (3.8.x, 3.9.x, 3.10.x, ...)

pip install -U lightning

Convert your PyTorch to Fabric

1. Create the Fabric object at the beginning of your training code.

   Copy

   import Lightning as L
   
   fabric = L.Fabric()
   

2. Call setup() on each model and optimizer pair and setup_dataloaders() on all your data loaders.

   Copy

   model, optimizer = fabric.setup(model, optimizer)
   dataloader = fabric.setup_dataloaders(dataloader)
   

3. Remove all .to and .cuda calls -> Fabric will take care of it.

   diffCopy

   - model.to(device)
   - batch.to(device)
   

4. Replace loss.backward() by fabric.backward(loss).

   diffCopy

   - loss.backward()
   + fabric.backward(loss)
   

5. Run the script from the terminal with

   bashCopy

   lightning run model path/to/train.py
   

or use the launch() method in a notebook. Learn more about launching distributed training.

FAQ

When to use Fabric?

• Minimum code changes- You want to scale your PyTorch model to use multi-GPU or use advanced strategies like DeepSpeed without having to refactor. You don’t care about structuring your code- you just want to scale it as fast as possible.
• Maximum control- Write your own training and/or inference logic down to the individual optimizer calls. You aren’t forced to conform to a standardized epoch-based training loop like the one in Lightning Trainer. You can do flexible iteration based training, meta-learning, cross-validation and other types of optimization algorithms without digging into framework internals. This also makes it super easy to adopt Fabric in existing PyTorch projects to speed-up and scale your models without the compromise on large refactors. Just remember: With great power comes a great responsibility.
• Maximum flexibility- You want to have full control over your entire training- in Fabric all features are opt-in, and it provides you with a tool box of primitives so you can build your own Trainer.

When to use the Lightning Trainer?

• You want to have all the engineering boilerplate handled for you - dozens of features like checkpointing, logging and early stopping out of the box. Less hassle, less error prone, easy to try different techniques and features.
• You want to have good defaults chosen for you - so you can have a better starting point.
• You want your code to be modular, readable and well structured - easy to share between projects and with collaborators.

Can I use Fabric with my LightningModule or Lightning Callback?

Yes :) Fabric works with PyTorch LightningModules and Callbacks, so you can choose how to structure your code and reuse existing models and callbacks as you wish. Read more here.

Examples

• GAN
• Meta learning
• Reinforcement learning
• K-Fold cross validation

Asking for help

If you have any questions please:

1. Read the docs.
2. Ask a question in our forum.
3. Join our discord community.