Shortcuts

LightningModule

A LightningModule organizes your PyTorch code into 6 sections:

  • Computations (init).

  • Train Loop (training_step)

  • Validation Loop (validation_step)

  • Test Loop (test_step)

  • Prediction Loop (predict_step)

  • Optimizers and LR Schedulers (configure_optimizers)



Notice a few things.

  1. It is the SAME code.

  2. The PyTorch code IS NOT abstracted - just organized.

  3. All the other code that’s not in the LightningModule has been automated for you by the Trainer.


net = Net()
trainer = Trainer()
trainer.fit(net)
  1. There are no .cuda() or .to(device) calls required. Lightning does these for you.


# don't do in Lightning
x = torch.Tensor(2, 3)
x = x.cuda()
x = x.to(device)

# do this instead
x = x  # leave it alone!

# or to init a new tensor
new_x = torch.Tensor(2, 3)
new_x = new_x.to(x)
  1. When running under a distributed strategy, Lightning handles the distributed sampler for you by default.


# Don't do in Lightning...
data = MNIST(...)
sampler = DistributedSampler(data)
DataLoader(data, sampler=sampler)

# do this instead
data = MNIST(...)
DataLoader(data)
  1. A LightningModule is a torch.nn.Module but with added functionality. Use it as such!


net = Net.load_from_checkpoint(PATH)
net.freeze()
out = net(x)

Thus, to use Lightning, you just need to organize your code which takes about 30 minutes, (and let’s be real, you probably should do anyway).


Starter Example

Here are the only required methods.

import pytorch_lightning as pl
import torch.nn as nn
import torch.nn.functional as F


class LitModel(pl.LightningModule):
    def __init__(self):
        super().__init__()
        self.l1 = nn.Linear(28 * 28, 10)

    def forward(self, x):
        return torch.relu(self.l1(x.view(x.size(0), -1)))

    def training_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self(x)
        loss = F.cross_entropy(y_hat, y)
        return loss

    def configure_optimizers(self):
        return torch.optim.Adam(self.parameters(), lr=0.02)

Which you can train by doing:

train_loader = DataLoader(MNIST(os.getcwd(), download=True, transform=transforms.ToTensor()))
trainer = pl.Trainer(max_epochs=1)
model = LitModel()

trainer.fit(model, train_dataloaders=train_loader)

The LightningModule has many convenience methods, but the core ones you need to know about are:

Name

Description

init

Define computations here

forward

Use for inference only (separate from training_step)

training_step

the complete training loop

validation_step

the complete validation loop

test_step

the complete test loop

predict_step

the complete prediction loop

configure_optimizers

define optimizers and LR schedulers


Training

Training Loop

To activate the training loop, override the training_step() method.

class LitClassifier(pl.LightningModule):
    def __init__(self, model):
        super().__init__()
        self.model = model

    def training_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self.model(x)
        loss = F.cross_entropy(y_hat, y)
        return loss

Under the hood, Lightning does the following (pseudocode):

# put model in train mode and enable gradient calculation
model.train()
torch.set_grad_enabled(True)

outs = []
for batch_idx, batch in enumerate(train_dataloader):
    loss = training_step(batch, batch_idx)
    outs.append(loss.detach())

    # clear gradients
    optimizer.zero_grad()

    # backward
    loss.backward()

    # update parameters
    optimizer.step()

Train Epoch-level Metrics

If you want to calculate epoch-level metrics and log them, use log().

def training_step(self, batch, batch_idx):
    x, y = batch
    y_hat = self.model(x)
    loss = F.cross_entropy(y_hat, y)

    # logs metrics for each training_step,
    # and the average across the epoch, to the progress bar and logger
    self.log("train_loss", loss, on_step=True, on_epoch=True, prog_bar=True, logger=True)
    return loss

The log() object automatically reduces the requested metrics across a complete epoch and devices. Here’s the pseudocode of what it does under the hood:

outs = []
for batch_idx, batch in enumerate(train_dataloader):
    # forward
    loss = training_step(batch, batch_idx)
    outs.append(loss)

    # clear gradients
    optimizer.zero_grad()

    # backward
    loss.backward()

    # update parameters
    optimizer.step()

epoch_metric = torch.mean(torch.stack([x for x in outs]))

Train Epoch-level Operations

If you need to do something with all the outputs of each training_step(), override the training_epoch_end() method.

def training_step(self, batch, batch_idx):
    x, y = batch
    y_hat = self.model(x)
    loss = F.cross_entropy(y_hat, y)
    preds = ...
    return {"loss": loss, "other_stuff": preds}


def training_epoch_end(self, training_step_outputs):
    all_preds = torch.stack(training_step_outputs)
    ...

The matching pseudocode is:

outs = []
for batch_idx, batch in enumerate(train_dataloader):
    # forward
    loss = training_step(batch, batch_idx)
    outs.append(loss)

    # clear gradients
    optimizer.zero_grad()

    # backward
    loss.backward()

    # update parameters
    optimizer.step()

training_epoch_end(outs)

Training with DataParallel

When training using a strategy that splits data from each batch across GPUs, sometimes you might need to aggregate them on the main GPU for processing (DP).

In this case, implement the training_step_end() method which will have outputs from all the devices and you can accumulate to get the effective results.

def training_step(self, batch, batch_idx):
    x, y = batch
    y_hat = self.model(x)
    loss = F.cross_entropy(y_hat, y)
    pred = ...
    return {"loss": loss, "pred": pred}


def training_step_end(self, batch_parts):
    # predictions from each GPU
    predictions = batch_parts["pred"]
    # losses from each GPU
    losses = batch_parts["loss"]

    gpu_0_prediction = predictions[0]
    gpu_1_prediction = predictions[1]

    # do something with both outputs
    return (losses[0] + losses[1]) / 2


def training_epoch_end(self, training_step_outputs):
    for out in training_step_outputs:
        ...

Here is the Lightning training pseudo-code for DP:

outs = []
for batch_idx, train_batch in enumerate(train_dataloader):
    batches = split_batch(train_batch)
    dp_outs = []
    for sub_batch in batches:
        # 1
        dp_out = training_step(sub_batch, batch_idx)
        dp_outs.append(dp_out)

    # 2
    out = training_step_end(dp_outs)
    outs.append(out)

# do something with the outputs for all batches
# 3
training_epoch_end(outs)

Validation

Validation Loop

To activate the validation loop while training, override the validation_step() method.

class LitModel(pl.LightningModule):
    def validation_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self.model(x)
        loss = F.cross_entropy(y_hat, y)
        self.log("val_loss", loss)

Under the hood, Lightning does the following (pseudocode):

# ...
for batch_idx, batch in enumerate(train_dataloader):
    loss = model.training_step(batch, batch_idx)
    loss.backward()
    # ...

    if validate_at_some_point:
        # disable grads + batchnorm + dropout
        torch.set_grad_enabled(False)
        model.eval()

        # ----------------- VAL LOOP ---------------
        for val_batch_idx, val_batch in enumerate(val_dataloader):
            val_out = model.validation_step(val_batch, val_batch_idx)
        # ----------------- VAL LOOP ---------------

        # enable grads + batchnorm + dropout
        torch.set_grad_enabled(True)
        model.train()

You can also run just the validation loop on your validation dataloaders by overriding validation_step() and calling validate().

model = Model()
trainer = Trainer()
trainer.validate(model)

Note

It is recommended to validate on single device to ensure each sample/batch gets evaluated exactly once. This is helpful to make sure benchmarking for research papers is done the right way. Otherwise, in a multi-device setting, samples could occur duplicated when DistributedSampler is used, for eg. with strategy="ddp". It replicates some samples on some devices to make sure all devices have same batch size in case of uneven inputs.

Validation Epoch-level Metrics

If you need to do something with all the outputs of each validation_step(), override the validation_epoch_end() method. Note that this method is called before training_epoch_end().

def validation_step(self, batch, batch_idx):
    x, y = batch
    y_hat = self.model(x)
    loss = F.cross_entropy(y_hat, y)
    pred = ...
    return pred


def validation_epoch_end(self, validation_step_outputs):
    all_preds = torch.stack(validation_step_outputs)
    ...

Validating with DataParallel

When validating using a strategy that splits data from each batch across GPUs, sometimes you might need to aggregate them on the main GPU for processing (DP).

In this case, implement the validation_step_end() method which will have outputs from all the devices and you can accumulate to get the effective results.

def validation_step(self, batch, batch_idx):
    x, y = batch
    y_hat = self.model(x)
    loss = F.cross_entropy(y_hat, y)
    pred = ...
    return {"loss": loss, "pred": pred}


def validation_step_end(self, batch_parts):
    # predictions from each GPU
    predictions = batch_parts["pred"]
    # losses from each GPU
    losses = batch_parts["loss"]

    gpu_0_prediction = predictions[0]
    gpu_1_prediction = predictions[1]

    # do something with both outputs
    return (losses[0] + losses[1]) / 2


def validation_epoch_end(self, validation_step_outputs):
    for out in validation_step_outputs:
        ...

Here is the Lightning validation pseudo-code for DP:

outs = []
for batch in dataloader:
    batches = split_batch(batch)
    dp_outs = []
    for sub_batch in batches:
        # 1
        dp_out = validation_step(sub_batch)
        dp_outs.append(dp_out)

    # 2
    out = validation_step_end(dp_outs)
    outs.append(out)

# do something with the outputs for all batches
# 3
validation_epoch_end(outs)

Testing

Test Loop

The process for enabling a test loop is the same as the process for enabling a validation loop. Please refer to the section above for details. For this you need to override the test_step() method.

The only difference is that the test loop is only called when test() is used.

model = Model()
trainer = Trainer()
trainer.fit(model)

# automatically loads the best weights for you
trainer.test(model)

There are two ways to call test():

# call after training
trainer = Trainer()
trainer.fit(model)

# automatically auto-loads the best weights from the previous run
trainer.test(dataloaders=test_dataloader)

# or call with pretrained model
model = MyLightningModule.load_from_checkpoint(PATH)
trainer = Trainer()
trainer.test(model, dataloaders=test_dataloader)

Note

It is recommended to validate on single device to ensure each sample/batch gets evaluated exactly once. This is helpful to make sure benchmarking for research papers is done the right way. Otherwise, in a multi-device setting, samples could occur duplicated when DistributedSampler is used, for eg. with strategy="ddp". It replicates some samples on some devices to make sure all devices have same batch size in case of uneven inputs.


Inference

Prediction Loop

By default, the predict_step() method runs the forward() method. In order to customize this behaviour, simply override the predict_step() method.

For the example let’s override predict_step and try out Monte Carlo Dropout:

class LitMCdropoutModel(pl.LightningModule):
    def __init__(self, model, mc_iteration):
        super().__init__()
        self.model = model
        self.dropout = nn.Dropout()
        self.mc_iteration = mc_iteration

    def predict_step(self, batch, batch_idx):
        # enable Monte Carlo Dropout
        self.dropout.train()

        # take average of `self.mc_iteration` iterations
        pred = torch.vstack([self.dropout(self.model(x)).unsqueeze(0) for _ in range(self.mc_iteration)]).mean(dim=0)
        return pred

Under the hood, Lightning does the following (pseudocode):

# disable grads + batchnorm + dropout
torch.set_grad_enabled(False)
model.eval()
all_preds = []

for batch_idx, batch in enumerate(predict_dataloader):
    pred = model.predict_step(batch, batch_idx)
    all_preds.append(pred)

There are two ways to call predict():

# call after training
trainer = Trainer()
trainer.fit(model)

# automatically auto-loads the best weights from the previous run
predictions = trainer.predict(dataloaders=predict_dataloader)

# or call with pretrained model
model = MyLightningModule.load_from_checkpoint(PATH)
trainer = Trainer()
predictions = trainer.predict(model, dataloaders=test_dataloader)

Inference in Research

If you want to perform inference with the system, you can add a forward method to the LightningModule.

Note

When using forward, you are responsible to call eval() and use the no_grad() context manager.

class Autoencoder(pl.LightningModule):
    def forward(self, x):
        return self.decoder(x)


model = Autoencoder()
model.eval()
with torch.no_grad():
    reconstruction = model(embedding)

The advantage of adding a forward is that in complex systems, you can do a much more involved inference procedure, such as text generation:

class Seq2Seq(pl.LightningModule):
    def forward(self, x):
        embeddings = self(x)
        hidden_states = self.encoder(embeddings)
        for h in hidden_states:
            # decode
            ...
        return decoded

In the case where you want to scale your inference, you should be using predict_step().

class Autoencoder(pl.LightningModule):
    def forward(self, x):
        return self.decoder(x)

    def predict_step(self, batch, batch_idx, dataloader_idx=0):
        # this calls forward
        return self(batch)


data_module = ...
model = Autoencoder()
trainer = Trainer(accelerator="gpu", devices=2)
trainer.predict(model, data_module)

Inference in Production

For cases like production, you might want to iterate different models inside a LightningModule.

from torchmetrics.functional import accuracy


class ClassificationTask(pl.LightningModule):
    def __init__(self, model):
        super().__init__()
        self.model = model

    def training_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self.model(x)
        loss = F.cross_entropy(y_hat, y)
        return loss

    def validation_step(self, batch, batch_idx):
        loss, acc = self._shared_eval_step(batch, batch_idx)
        metrics = {"val_acc": acc, "val_loss": loss}
        self.log_dict(metrics)
        return metrics

    def test_step(self, batch, batch_idx):
        loss, acc = self._shared_eval_step(batch, batch_idx)
        metrics = {"test_acc": acc, "test_loss": loss}
        self.log_dict(metrics)
        return metrics

    def _shared_eval_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self.model(x)
        loss = F.cross_entropy(y_hat, y)
        acc = accuracy(y_hat, y)
        return loss, acc

    def predict_step(self, batch, batch_idx, dataloader_idx=0):
        x, y = batch
        y_hat = self.model(x)
        return y_hat

    def configure_optimizers(self):
        return torch.optim.Adam(self.model.parameters(), lr=0.02)

Then pass in any arbitrary model to be fit with this task

for model in [resnet50(), vgg16(), BidirectionalRNN()]:
    task = ClassificationTask(model)

    trainer = Trainer(accelerator="gpu", devices=2)
    trainer.fit(task, train_dataloaders=train_dataloader, val_dataloaders=val_dataloader)

Tasks can be arbitrarily complex such as implementing GAN training, self-supervised or even RL.

class GANTask(pl.LightningModule):
    def __init__(self, generator, discriminator):
        super().__init__()
        self.generator = generator
        self.discriminator = discriminator

    ...

When used like this, the model can be separated from the Task and thus used in production without needing to keep it in a LightningModule.

The following example shows how you can run inference in the Python runtime:

task = ClassificationTask(model)
trainer = Trainer(accelerator="gpu", devices=2)
trainer.fit(task, train_dataloader, val_dataloader)
trainer.save_checkpoint("best_model.ckpt")

# use model after training or load weights and drop into the production system
model = ClassificationTask.load_from_checkpoint("best_model.ckpt")
x = ...
model.eval()
with torch.no_grad():
    y_hat = model(x)

Check out Inference in Production guide to learn about the possible ways to perform inference in production.


Save Hyperparameters

Often times we train many versions of a model. You might share that model or come back to it a few months later at which point it is very useful to know how that model was trained (i.e.: what learning rate, neural network, etc…).

Lightning has a standardized way of saving the information for you in checkpoints and YAML files. The goal here is to improve readability and reproducibility.

save_hyperparameters

Use save_hyperparameters() within your LightningModule’s __init__ method. It will enable Lightning to store all the provided arguments under the self.hparams attribute. These hyperparameters will also be stored within the model checkpoint, which simplifies model re-instantiation after training.

class LitMNIST(LightningModule):
    def __init__(self, layer_1_dim=128, learning_rate=1e-2):
        super().__init__()
        # call this to save (layer_1_dim=128, learning_rate=1e-4) to the checkpoint
        self.save_hyperparameters()

        # equivalent
        self.save_hyperparameters("layer_1_dim", "learning_rate")

        # Now possible to access layer_1_dim from hparams
        self.hparams.layer_1_dim

In addition, loggers that support it will automatically log the contents of self.hparams.

Excluding hyperparameters

By default, every parameter of the __init__ method will be considered a hyperparameter to the LightningModule. However, sometimes some parameters need to be excluded from saving, for example when they are not serializable. Those parameters should be provided back when reloading the LightningModule. In this case, exclude them explicitly:

class LitMNIST(LightningModule):
    def __init__(self, loss_fx, generator_network, layer_1_dim=128):
        super().__init__()
        self.layer_1_dim = layer_1_dim
        self.loss_fx = loss_fx

        # call this to save only (layer_1_dim=128) to the checkpoint
        self.save_hyperparameters("layer_1_dim")

        # equivalent
        self.save_hyperparameters(ignore=["loss_fx", "generator_network"])

load_from_checkpoint

LightningModules that have hyperparameters automatically saved with save_hyperparameters() can conveniently be loaded and instantiated directly from a checkpoint with load_from_checkpoint():

# to load specify the other args
model = LitMNIST.load_from_checkpoint(PATH, loss_fx=torch.nn.SomeOtherLoss, generator_network=MyGenerator())

If parameters were excluded, they need to be provided at the time of loading:

# the excluded parameters were `loss_fx` and `generator_network`
model = LitMNIST.load_from_checkpoint(PATH, loss_fx=torch.nn.SomeOtherLoss, generator_network=MyGenerator())

Child Modules

Research projects tend to test different approaches to the same dataset. This is very easy to do in Lightning with inheritance.

For example, imagine we now want to train an AutoEncoder to use as a feature extractor for images. The only things that change in the LitAutoEncoder model are the init, forward, training, validation and test step.

class Encoder(torch.nn.Module):
    ...


class Decoder(torch.nn.Module):
    ...


class AutoEncoder(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = Encoder()
        self.decoder = Decoder()

    def forward(self, x):
        return self.decoder(self.encoder(x))


class LitAutoEncoder(LightningModule):
    def __init__(self, auto_encoder):
        super().__init__()
        self.auto_encoder = auto_encoder
        self.metric = torch.nn.MSELoss()

    def forward(self, x):
        return self.auto_encoder.encoder(x)

    def training_step(self, batch, batch_idx):
        x, _ = batch
        x_hat = self.auto_encoder(x)
        loss = self.metric(x, x_hat)
        return loss

    def validation_step(self, batch, batch_idx):
        self._shared_eval(batch, batch_idx, "val")

    def test_step(self, batch, batch_idx):
        self._shared_eval(batch, batch_idx, "test")

    def _shared_eval(self, batch, batch_idx, prefix):
        x, _ = batch
        x_hat = self.auto_encoder(x)
        loss = self.metric(x, x_hat)
        self.log(f"{prefix}_loss", loss)

and we can train this using the Trainer:

auto_encoder = AutoEncoder()
lightning_module = LitAutoEncoder(auto_encoder)
trainer = Trainer()
trainer.fit(lightning_module, train_dataloader, val_dataloader)

And remember that the forward method should define the practical use of a LightningModule. In this case, we want to use the LitAutoEncoder to extract image representations:

some_images = torch.Tensor(32, 1, 28, 28)
representations = lightning_module(some_images)

LightningModule API

Methods

all_gather

LightningModule.all_gather(data, group=None, sync_grads=False)[source]

Allows users to call self.all_gather() from the LightningModule, thus making the all_gather operation accelerator agnostic. all_gather is a function provided by accelerators to gather a tensor from several distributed processes.

Parameters
  • data (Union[Tensor, Dict, List, Tuple]) – int, float, tensor of shape (batch, …), or a (possibly nested) collection thereof.

  • group (Optional[Any]) – the process group to gather results from. Defaults to all processes (world)

  • sync_grads (bool) – flag that allows users to synchronize gradients for the all_gather operation

Return type

Union[Tensor, Dict, List, Tuple]

Returns

A tensor of shape (world_size, batch, …), or if the input was a collection the output will also be a collection with tensors of this shape.

configure_callbacks

LightningModule.configure_callbacks()[source]

Configure model-specific callbacks. When the model gets attached, e.g., when .fit() or .test() gets called, the list or a callback returned here will be merged with the list of callbacks passed to the Trainer’s callbacks argument. If a callback returned here has the same type as one or several callbacks already present in the Trainer’s callbacks list, it will take priority and replace them. In addition, Lightning will make sure ModelCheckpoint callbacks run last.

Return type

Union[Sequence[Callback], Callback]

Returns

A callback or a list of callbacks which will extend the list of callbacks in the Trainer.

Example:

def configure_callbacks(self):
    early_stop = EarlyStopping(monitor="val_acc", mode="max")
    checkpoint = ModelCheckpoint(monitor="val_loss")
    return [early_stop, checkpoint]

configure_optimizers

LightningModule.configure_optimizers()[source]

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple.

Return type

Any

Returns

Any of these 6 options.

  • Single optimizer.

  • List or Tuple of optimizers.

  • Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

  • Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

  • Tuple of dictionaries as described above, with an optional "frequency" key.

  • None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

# The ReduceLROnPlateau scheduler requires a monitor
def configure_optimizers(self):
    optimizer = Adam(...)
    return {
        "optimizer": optimizer,
        "lr_scheduler": {
            "scheduler": ReduceLROnPlateau(optimizer, ...),
            "monitor": "metric_to_track",
            "frequency": "indicates how often the metric is updated"
            # If "monitor" references validation metrics, then "frequency" should be set to a
            # multiple of "trainer.check_val_every_n_epoch".
        },
    }


# In the case of two optimizers, only one using the ReduceLROnPlateau scheduler
def configure_optimizers(self):
    optimizer1 = Adam(...)
    optimizer2 = SGD(...)
    scheduler1 = ReduceLROnPlateau(optimizer1, ...)
    scheduler2 = LambdaLR(optimizer2, ...)
    return (
        {
            "optimizer": optimizer1,
            "lr_scheduler": {
                "scheduler": scheduler1,
                "monitor": "metric_to_track",
            },
        },
        {"optimizer": optimizer2, "lr_scheduler": scheduler2},
    )

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

The frequency value specified in a dict along with the optimizer key is an int corresponding to the number of sequential batches optimized with the specific optimizer. It should be given to none or to all of the optimizers. There is a difference between passing multiple optimizers in a list, and passing multiple optimizers in dictionaries with a frequency of 1:

  • In the former case, all optimizers will operate on the given batch in each optimization step.

  • In the latter, only one optimizer will operate on the given batch at every step.

This is different from the frequency value specified in the lr_scheduler_config mentioned above.

def configure_optimizers(self):
    optimizer_one = torch.optim.SGD(self.model.parameters(), lr=0.01)
    optimizer_two = torch.optim.SGD(self.model.parameters(), lr=0.01)
    return [
        {"optimizer": optimizer_one, "frequency": 5},
        {"optimizer": optimizer_two, "frequency": 10},
    ]

In this example, the first optimizer will be used for the first 5 steps, the second optimizer for the next 10 steps and that cycle will continue. If an LR scheduler is specified for an optimizer using the lr_scheduler key in the above dict, the scheduler will only be updated when its optimizer is being used.

Examples:

# most cases. no learning rate scheduler
def configure_optimizers(self):
    return Adam(self.parameters(), lr=1e-3)

# multiple optimizer case (e.g.: GAN)
def configure_optimizers(self):
    gen_opt = Adam(self.model_gen.parameters(), lr=0.01)
    dis_opt = Adam(self.model_dis.parameters(), lr=0.02)
    return gen_opt, dis_opt

# example with learning rate schedulers
def configure_optimizers(self):
    gen_opt = Adam(self.model_gen.parameters(), lr=0.01)
    dis_opt = Adam(self.model_dis.parameters(), lr=0.02)
    dis_sch = CosineAnnealing(dis_opt, T_max=10)
    return [gen_opt, dis_opt], [dis_sch]

# example with step-based learning rate schedulers
# each optimizer has its own scheduler
def configure_optimizers(self):
    gen_opt = Adam(self.model_gen.parameters(), lr=0.01)
    dis_opt = Adam(self.model_dis.parameters(), lr=0.02)
    gen_sch = {
        'scheduler': ExponentialLR(gen_opt, 0.99),
        'interval': 'step'  # called after each training step
    }
    dis_sch = CosineAnnealing(dis_opt, T_max=10) # called every epoch
    return [gen_opt, dis_opt], [gen_sch, dis_sch]

# example with optimizer frequencies
# see training procedure in `Improved Training of Wasserstein GANs`, Algorithm 1
# https://arxiv.org/abs/1704.00028
def configure_optimizers(self):
    gen_opt = Adam(self.model_gen.parameters(), lr=0.01)
    dis_opt = Adam(self.model_dis.parameters(), lr=0.02)
    n_critic = 5
    return (
        {'optimizer': dis_opt, 'frequency': n_critic},
        {'optimizer': gen_opt, 'frequency': 1}
    )

Note

Some things to know:

  • Lightning calls .backward() and .step() on each optimizer as needed.

  • If learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

  • If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizers.

  • If you use multiple optimizers, training_step() will have an additional optimizer_idx parameter.

  • If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

  • If you use multiple optimizers, gradients will be calculated only for the parameters of current optimizer at each training step.

  • If you need to control how often those optimizers step or override the default .step() schedule, override the optimizer_step() hook.

forward

LightningModule.forward(*args, **kwargs)[source]

Same as torch.nn.Module.forward().

Parameters
  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

Return type

Any

Returns

Your model’s output

freeze

LightningModule.freeze()[source]

Freeze all params for inference.

Example:

model = MyLightningModule(...)
model.freeze()
Return type

None

log

LightningModule.log(name, value, prog_bar=False, logger=None, on_step=None, on_epoch=None, reduce_fx='mean', enable_graph=False, sync_dist=False, sync_dist_group=None, add_dataloader_idx=True, batch_size=None, metric_attribute=None, rank_zero_only=False)[source]

Log a key, value pair.

Example:

self.log('train_loss', loss)

The default behavior per hook is documented here: Automatic Logging.

Parameters
  • name (str) – key to log.

  • value (Union[Metric, Tensor, int, float, Mapping[str, Union[Metric, Tensor, int, float]]]) – value to log. Can be a float, Tensor, Metric, or a dictionary of the former.

  • prog_bar (bool) – if True logs to the progress bar.

  • logger (Optional[bool]) – if True logs to the logger.

  • on_step (Optional[bool]) – if True logs at this step. The default value is determined by the hook. See Automatic Logging for details.

  • on_epoch (Optional[bool]) – if True logs epoch accumulated metrics. The default value is determined by the hook. See Automatic Logging for details.

  • reduce_fx (Union[str, Callable]) – reduction function over step values for end of epoch. torch.mean() by default.

  • enable_graph (bool) – if True, will not auto detach the graph.

  • sync_dist (bool) – if True, reduces the metric across devices. Use with care as this may lead to a significant communication overhead.

  • sync_dist_group (Optional[Any]) – the DDP group to sync across.

  • add_dataloader_idx (bool) – if True, appends the index of the current dataloader to the name (when using multiple dataloaders). If False, user needs to give unique names for each dataloader to not mix the values.

  • batch_size (Optional[int]) – Current batch_size. This will be directly inferred from the loaded batch, but for some data structures you might need to explicitly provide it.

  • metric_attribute (Optional[str]) – To restore the metric state, Lightning requires the reference of the torchmetrics.Metric in your model. This is found automatically if it is a model attribute.

  • rank_zero_only (bool) – Whether the value will be logged only on rank 0. This will prevent synchronization which would produce a deadlock as not all processes would perform this log call.

Return type

None

log_dict

LightningModule.log_dict(dictionary, prog_bar=False, logger=None, on_step=None, on_epoch=None, reduce_fx='mean', enable_graph=False, sync_dist=False, sync_dist_group=None, add_dataloader_idx=True, batch_size=None, rank_zero_only=False)[source]

Log a dictionary of values at once.

Example:

values = {'loss': loss, 'acc': acc, ..., 'metric_n': metric_n}
self.log_dict(values)
Parameters
  • dictionary (Mapping[str, Union[Metric, Tensor, int, float, Mapping[str, Union[Metric, Tensor, int, float]]]]) – key value pairs. The values can be a float, Tensor, Metric, a dictionary of the former or a MetricCollection.

  • prog_bar (bool) – if True logs to the progress base.

  • logger (Optional[bool]) – if True logs to the logger.

  • on_step (Optional[bool]) – if True logs at this step. None auto-logs for training_step but not validation/test_step. The default value is determined by the hook. See Automatic Logging for details.

  • on_epoch (Optional[bool]) – if True logs epoch accumulated metrics. None auto-logs for val/test step but not training_step. The default value is determined by the hook. See Automatic Logging for details.

  • reduce_fx (Union[str, Callable]) – reduction function over step values for end of epoch. torch.mean() by default.

  • enable_graph (bool) – if True, will not auto-detach the graph

  • sync_dist (bool) – if True, reduces the metric across GPUs/TPUs. Use with care as this may lead to a significant communication overhead.

  • sync_dist_group (Optional[Any]) – the ddp group to sync across.

  • add_dataloader_idx (bool) – if True, appends the index of the current dataloader to the name (when using multiple). If False, user needs to give unique names for each dataloader to not mix values.

  • batch_size (Optional[int]) – Current batch size. This will be directly inferred from the loaded batch, but some data structures might need to explicitly provide it.

  • rank_zero_only (bool) – Whether the value will be logged only on rank 0. This will prevent synchronization which would produce a deadlock as not all processes would perform this log call.

Return type

None

lr_schedulers

LightningModule.lr_schedulers()[source]

Returns the learning rate scheduler(s) that are being used during training. Useful for manual optimization.

Return type

Union[None, List[Union[LRScheduler, ReduceLROnPlateau]], LRScheduler, ReduceLROnPlateau]

Returns

A single scheduler, or a list of schedulers in case multiple ones are present, or None if no schedulers were returned in configure_optimizers().

manual_backward

LightningModule.manual_backward(loss, *args, **kwargs)[source]

Call this directly from your training_step() when doing optimizations manually. By using this, Lightning can ensure that all the proper scaling gets applied when using mixed precision.

See manual optimization for more examples.

Example:

def training_step(...):
    opt = self.optimizers()
    loss = ...
    opt.zero_grad()
    # automatically applies scaling, etc...
    self.manual_backward(loss)
    opt.step()
Parameters
  • loss (Tensor) – The tensor on which to compute gradients. Must have a graph attached.

  • *args – Additional positional arguments to be forwarded to backward()

  • **kwargs – Additional keyword arguments to be forwarded to backward()

Return type

None

optimizers

LightningModule.optimizers(use_pl_optimizer: Literal[True] = True) Union[pytorch_lightning.core.optimizer.LightningOptimizer, List[pytorch_lightning.core.optimizer.LightningOptimizer]][source]
LightningModule.optimizers(use_pl_optimizer: Literal[False]) Union[torch.optim.optimizer.Optimizer, List[torch.optim.optimizer.Optimizer]]
LightningModule.optimizers(use_pl_optimizer: bool) Union[torch.optim.optimizer.Optimizer, pytorch_lightning.core.optimizer.LightningOptimizer, lightning_fabric.wrappers._FabricOptimizer, List[torch.optim.optimizer.Optimizer], List[pytorch_lightning.core.optimizer.LightningOptimizer], List[lightning_fabric.wrappers._FabricOptimizer]]

Returns the optimizer(s) that are being used during training. Useful for manual optimization.

Parameters

use_pl_optimizer (bool) – If True, will wrap the optimizer(s) in a LightningOptimizer for automatic handling of precision and profiling.

Return type

Union[Optimizer, LightningOptimizer, _FabricOptimizer, List[Optimizer], List[LightningOptimizer], List[_FabricOptimizer]]

Returns

A single optimizer, or a list of optimizers in case multiple ones are present.

print

LightningModule.print(*args, **kwargs)[source]

Prints only from process 0. Use this in any distributed mode to log only once.

Parameters
  • *args – The thing to print. The same as for Python’s built-in print function.

  • **kwargs – The same as for Python’s built-in print function.

Example:

def forward(self, x):
    self.print(x, 'in forward')
Return type

None

predict_step

LightningModule.predict_step(batch, batch_idx, dataloader_idx=0)[source]

Step function called during predict(). By default, it calls forward(). Override to add any processing logic.

The predict_step() is used to scale inference on multi-devices.

To prevent an OOM error, it is possible to use BasePredictionWriter callback to write the predictions to disk or database after each batch or on epoch end.

The BasePredictionWriter should be used while using a spawn based accelerator. This happens for Trainer(strategy="ddp_spawn") or training on 8 TPU cores with Trainer(accelerator="tpu", devices=8) as predictions won’t be returned.

Example

class MyModel(LightningModule):

    def predict_step(self, batch, batch_idx, dataloader_idx=0):
        return self(batch)

dm = ...
model = MyModel()
trainer = Trainer(accelerator="gpu", devices=2)
predictions = trainer.predict(model, dm)
Parameters
  • batch (Any) – Current batch.

  • batch_idx (int) – Index of current batch.

  • dataloader_idx (int) – Index of the current dataloader.

Return type

Any

Returns

Predicted output

save_hyperparameters

LightningModule.save_hyperparameters(*args, ignore=None, frame=None, logger=True)

Save arguments to hparams attribute.

Parameters
  • args (Any) – single object of dict, NameSpace or OmegaConf or string names or arguments from class __init__

  • ignore (Union[Sequence[str], str, None]) – an argument name or a list of argument names from class __init__ to be ignored

  • frame (Optional[frame]) – a frame object. Default is None

  • logger (bool) – Whether to send the hyperparameters to the logger. Default: True

Example::
>>> from pytorch_lightning.core.mixins import HyperparametersMixin
>>> class ManuallyArgsModel(HyperparametersMixin):
...     def __init__(self, arg1, arg2, arg3):
...         super().__init__()
...         # manually assign arguments
...         self.save_hyperparameters('arg1', 'arg3')
...     def forward(self, *args, **kwargs):
...         ...
>>> model = ManuallyArgsModel(1, 'abc', 3.14)
>>> model.hparams
"arg1": 1
"arg3": 3.14
>>> from pytorch_lightning.core.mixins import HyperparametersMixin
>>> class AutomaticArgsModel(HyperparametersMixin):
...     def __init__(self, arg1, arg2, arg3):
...         super().__init__()
...         # equivalent automatic
...         self.save_hyperparameters()
...     def forward(self, *args, **kwargs):
...         ...
>>> model = AutomaticArgsModel(1, 'abc', 3.14)
>>> model.hparams
"arg1": 1
"arg2": abc
"arg3": 3.14
>>> from pytorch_lightning.core.mixins import HyperparametersMixin
>>> class SingleArgModel(HyperparametersMixin):
...     def __init__(self, params):
...         super().__init__()
...         # manually assign single argument
...         self.save_hyperparameters(params)
...     def forward(self, *args, **kwargs):
...         ...
>>> model = SingleArgModel(Namespace(p1=1, p2='abc', p3=3.14))
>>> model.hparams
"p1": 1
"p2": abc
"p3": 3.14
>>> from pytorch_lightning.core.mixins import HyperparametersMixin
>>> class ManuallyArgsModel(HyperparametersMixin):
...     def __init__(self, arg1, arg2, arg3):
...         super().__init__()
...         # pass argument(s) to ignore as a string or in a list
...         self.save_hyperparameters(ignore='arg2')
...     def forward(self, *args, **kwargs):
...         ...
>>> model = ManuallyArgsModel(1, 'abc', 3.14)
>>> model.hparams
"arg1": 1
"arg3": 3.14
Return type

None

toggle_optimizer

LightningModule.toggle_optimizer(optimizer, optimizer_idx)[source]

Makes sure only the gradients of the current optimizer’s parameters are calculated in the training step to prevent dangling gradients in multiple-optimizer setup.

This is only called automatically when automatic optimization is enabled and multiple optimizers are used. It works with untoggle_optimizer() to make sure param_requires_grad_state is properly reset.

Parameters
Return type

None

test_step

LightningModule.test_step(*args, **kwargs)[source]

Operates on a single batch of data from the test set. In this step you’d normally generate examples or calculate anything of interest such as accuracy.

# the pseudocode for these calls
test_outs = []
for test_batch in test_data:
    out = test_step(test_batch)
    test_outs.append(out)
test_epoch_end(test_outs)
Parameters
  • batch – The output of your DataLoader.

  • batch_idx – The index of this batch.

  • dataloader_id – The index of the dataloader that produced this batch. (only if multiple test dataloaders used).

Return type

Union[Tensor, Dict[str, Any], None]

Returns

Any of.

  • Any object or value

  • None - Testing will skip to the next batch

# if you have one test dataloader:
def test_step(self, batch, batch_idx):
    ...


# if you have multiple test dataloaders:
def test_step(self, batch, batch_idx, dataloader_idx=0):
    ...

Examples:

# CASE 1: A single test dataset
def test_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    test_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'test_loss': loss, 'test_acc': test_acc})

If you pass in multiple test dataloaders, test_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    ...

Note

If you don’t need to test you don’t need to implement this method.

Note

When the test_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of the test epoch, the model goes back to training mode and gradients are enabled.

test_step_end

LightningModule.test_step_end(*args, **kwargs)[source]

Use this when testing with DP because test_step() will operate on only part of the batch. However, this is still optional and only needed for things like softmax or NCE loss.

Note

If you later switch to ddp or some other mode, this will still be called so that you don’t have to change your code.

# pseudocode
sub_batches = split_batches_for_dp(batch)
step_output = [test_step(sub_batch) for sub_batch in sub_batches]
test_step_end(step_output)
Parameters

step_output – What you return in test_step() for each batch part.

Return type

Union[Tensor, Dict[str, Any], None]

Returns

None or anything

# WITHOUT test_step_end
# if used in DP, this batch is 1/num_gpus large
def test_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self(x)
    loss = self.softmax(out)
    self.log("test_loss", loss)


# --------------
# with test_step_end to do softmax over the full batch
def test_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self.encoder(x)
    return out


def test_step_end(self, output_results):
    # this out is now the full size of the batch
    all_test_step_outs = output_results.out
    loss = nce_loss(all_test_step_outs)
    self.log("test_loss", loss)

See also

See the Multi GPU Training guide for more details.

test_epoch_end

LightningModule.test_epoch_end(outputs)[source]

Called at the end of a test epoch with the output of all test steps.

# the pseudocode for these calls
test_outs = []
for test_batch in test_data:
    out = test_step(test_batch)
    test_outs.append(out)
test_epoch_end(test_outs)
Parameters

outputs (Union[List[Union[Tensor, Dict[str, Any]]], List[List[Union[Tensor, Dict[str, Any]]]]]) – List of outputs you defined in test_step_end(), or if there are multiple dataloaders, a list containing a list of outputs for each dataloader

Return type

None

Returns

None

Note

If you didn’t define a test_step(), this won’t be called.

Examples

With a single dataloader:

def test_epoch_end(self, outputs):
    # do something with the outputs of all test batches
    all_test_preds = test_step_outputs.predictions

    some_result = calc_all_results(all_test_preds)
    self.log(some_result)

With multiple dataloaders, outputs will be a list of lists. The outer list contains one entry per dataloader, while the inner list contains the individual outputs of each test step for that dataloader.

def test_epoch_end(self, outputs):
    final_value = 0
    for dataloader_outputs in outputs:
        for test_step_out in dataloader_outputs:
            # do something
            final_value += test_step_out

    self.log("final_metric", final_value)

to_onnx

LightningModule.to_onnx(file_path, input_sample=None, **kwargs)[source]

Saves the model in ONNX format.

Parameters
  • file_path (Union[str, Path]) – The path of the file the onnx model should be saved to.

  • input_sample (Optional[Any]) – An input for tracing. Default: None (Use self.example_input_array)

  • **kwargs – Will be passed to torch.onnx.export function.

Example:

class SimpleModel(LightningModule):
    def __init__(self):
        super().__init__()
        self.l1 = torch.nn.Linear(in_features=64, out_features=4)

    def forward(self, x):
        return torch.relu(self.l1(x.view(x.size(0), -1)))

model = SimpleModel()
model.to_onnx("my-model.onnx", torch.randn((1, 64)), export_params=True)
Return type

None

to_torchscript

LightningModule.to_torchscript(file_path=None, method='script', example_inputs=None, **kwargs)[source]

By default compiles the whole model to a ScriptModule. If you want to use tracing, please provided the argument method='trace' and make sure that either the example_inputs argument is provided, or the model has example_input_array set. If you would like to customize the modules that are scripted you should override this method. In case you want to return multiple modules, we recommend using a dictionary.

Parameters
  • file_path (Union[str, Path, None]) – Path where to save the torchscript. Default: None (no file saved).

  • method (Optional[str]) – Whether to use TorchScript’s script or trace method. Default: ‘script’

  • example_inputs (Optional[Any]) – An input to be used to do tracing when method is set to ‘trace’. Default: None (uses example_input_array)

  • **kwargs – Additional arguments that will be passed to the torch.jit.script() or torch.jit.trace() function.

Note

  • Requires the implementation of the forward() method.

  • The exported script will be set to evaluation mode.

  • It is recommended that you install the latest supported version of PyTorch to use this feature without limitations. See also the torch.jit documentation for supported features.

Example:

class SimpleModel(LightningModule):
    def __init__(self):
        super().__init__()
        self.l1 = torch.nn.Linear(in_features=64, out_features=4)

    def forward(self, x):
       return torch.relu(self.l1(x.view(x.size(0), -1)))

model = SimpleModel()
model.to_torchscript(file_path="model.pt")

torch.jit.save(model.to_torchscript(
    file_path="model_trace.pt", method='trace', example_inputs=torch.randn(1, 64))
)
Return type

Union[ScriptModule, Dict[str, ScriptModule]]

Returns

This LightningModule as a torchscript, regardless of whether file_path is defined or not.

training_step

LightningModule.training_step(*args, **kwargs)[source]

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters
  • batch (Tensor | (Tensor, …) | [Tensor, …]) – The output of your DataLoader. A tensor, tuple or list.

  • batch_idx (int) – Integer displaying index of this batch

  • optimizer_idx (int) – When using multiple optimizers, this argument will also be present.

  • hiddens (Any) – Passed in if truncated_bptt_steps > 0.

Return type

Union[Tensor, Dict[str, Any]]

Returns

Any of.

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'

  • None - Training will skip to the next batch. This is only for automatic optimization.

    This is not supported for multi-GPU, TPU, IPU, or DeepSpeed.

In this step you’d normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

If you define multiple optimizers, this step will be called with an additional optimizer_idx parameter.

# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx, optimizer_idx):
    if optimizer_idx == 0:
        # do training_step with encoder
        ...
    if optimizer_idx == 1:
        # do training_step with decoder
        ...

If you add truncated back propagation through time you will also get an additional argument with the hidden states of the previous step.

# Truncated back-propagation through time
def training_step(self, batch, batch_idx, hiddens):
    # hiddens are the hidden states from the previous truncated backprop step
    out, hiddens = self.lstm(data, hiddens)
    loss = ...
    return {"loss": loss, "hiddens": hiddens}

Note

The loss value shown in the progress bar is smoothed (averaged) over the last values, so it differs from the actual loss returned in train/validation step.

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

training_step_end

LightningModule.training_step_end(step_output)[source]

Use this when training with dp because training_step() will operate on only part of the batch. However, this is still optional and only needed for things like softmax or NCE loss.

Note

If you later switch to ddp or some other mode, this will still be called so that you don’t have to change your code

# pseudocode
sub_batches = split_batches_for_dp(batch)
step_output = [training_step(sub_batch) for sub_batch in sub_batches]
training_step_end(step_output)
Parameters

step_output (Union[Tensor, Dict[str, Any]]) – What you return in training_step for each batch part.

Return type

Union[Tensor, Dict[str, Any]]

Returns

Anything

When using the DP strategy, only a portion of the batch is inside the training_step:

def training_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self(x)

    # softmax uses only a portion of the batch in the denominator
    loss = self.softmax(out)
    loss = nce_loss(loss)
    return loss

If you wish to do something with all the parts of the batch, then use this method to do it:

def training_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self.encoder(x)
    return {"pred": out}


def training_step_end(self, training_step_outputs):
    gpu_0_pred = training_step_outputs[0]["pred"]
    gpu_1_pred = training_step_outputs[1]["pred"]
    gpu_n_pred = training_step_outputs[n]["pred"]

    # this softmax now uses the full batch
    loss = nce_loss([gpu_0_pred, gpu_1_pred, gpu_n_pred])
    return loss

See also

See the Multi GPU Training guide for more details.

training_epoch_end

LightningModule.training_epoch_end(outputs)[source]

Called at the end of the training epoch with the outputs of all training steps. Use this in case you need to do something with all the outputs returned by training_step().

# the pseudocode for these calls
train_outs = []
for train_batch in train_data:
    out = training_step(train_batch)
    train_outs.append(out)
training_epoch_end(train_outs)
Parameters

outputs (List[Union[Tensor, Dict[str, Any]]]) – List of outputs you defined in training_step(). If there are multiple optimizers or when using truncated_bptt_steps > 0, the lists have the dimensions (n_batches, tbptt_steps, n_optimizers). Dimensions of length 1 are squeezed.

Return type

None

Returns

None

Note

If this method is not overridden, this won’t be called.

def training_epoch_end(self, training_step_outputs):
    # do something with all training_step outputs
    for out in training_step_outputs:
        ...

unfreeze

LightningModule.unfreeze()[source]

Unfreeze all parameters for training.

model = MyLightningModule(...)
model.unfreeze()
Return type

None

untoggle_optimizer

LightningModule.untoggle_optimizer(optimizer_idx)[source]

Resets the state of required gradients that were toggled with toggle_optimizer().

This is only called automatically when automatic optimization is enabled and multiple optimizers are used.

Parameters

optimizer_idx (int) – The index of the optimizer to untoggle.

Return type

None

validation_step

LightningModule.validation_step(*args, **kwargs)[source]

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

# the pseudocode for these calls
val_outs = []
for val_batch in val_data:
    out = validation_step(val_batch)
    val_outs.append(out)
validation_epoch_end(val_outs)
Parameters
  • batch – The output of your DataLoader.

  • batch_idx – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple val dataloaders used)

Return type

Union[Tensor, Dict[str, Any], None]

Returns

  • Any object or value

  • None - Validation will skip to the next batch

# pseudocode of order
val_outs = []
for val_batch in val_data:
    out = validation_step(val_batch)
    if defined("validation_step_end"):
        out = validation_step_end(out)
    val_outs.append(out)
val_outs = validation_epoch_end(val_outs)
# if you have one val dataloader:
def validation_step(self, batch, batch_idx):
    ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    ...

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

validation_step_end

LightningModule.validation_step_end(*args, **kwargs)[source]

Use this when validating with dp because validation_step() will operate on only part of the batch. However, this is still optional and only needed for things like softmax or NCE loss.

Note

If you later switch to ddp or some other mode, this will still be called so that you don’t have to change your code.

# pseudocode
sub_batches = split_batches_for_dp(batch)
step_output = [validation_step(sub_batch) for sub_batch in sub_batches]
validation_step_end(step_output)
Parameters

step_output – What you return in validation_step() for each batch part.

Return type

Union[Tensor, Dict[str, Any], None]

Returns

None or anything

# WITHOUT validation_step_end
# if used in DP, this batch is 1/num_gpus large
def validation_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self.encoder(x)
    loss = self.softmax(out)
    loss = nce_loss(loss)
    self.log("val_loss", loss)


# --------------
# with validation_step_end to do softmax over the full batch
def validation_step(self, batch, batch_idx):
    # batch is 1/num_gpus big
    x, y = batch

    out = self(x)
    return out


def validation_step_end(self, val_step_outputs):
    for out in val_step_outputs:
        ...

See also

See the Multi GPU Training guide for more details.

validation_epoch_end

LightningModule.validation_epoch_end(outputs)[source]

Called at the end of the validation epoch with the outputs of all validation steps.

# the pseudocode for these calls
val_outs = []
for val_batch in val_data:
    out = validation_step(val_batch)
    val_outs.append(out)
validation_epoch_end(val_outs)
Parameters

outputs (Union[List[Union[Tensor, Dict[str, Any]]], List[List[Union[Tensor, Dict[str, Any]]]]]) – List of outputs you defined in validation_step(), or if there are multiple dataloaders, a list containing a list of outputs for each dataloader.

Return type

None

Returns

None

Note

If you didn’t define a validation_step(), this won’t be called.

Examples

With a single dataloader:

def validation_epoch_end(self, val_step_outputs):
    for out in val_step_outputs:
        ...

With multiple dataloaders, outputs will be a list of lists. The outer list contains one entry per dataloader, while the inner list contains the individual outputs of each validation step for that dataloader.

def validation_epoch_end(self, outputs):
    for dataloader_output_result in outputs:
        dataloader_outs = dataloader_output_result.dataloader_i_outputs

    self.log("final_metric", final_value)

Properties

These are properties available in a LightningModule.

current_epoch

The number of epochs run.

def training_step(self, batch, batch_idx):
    if self.current_epoch == 0:
        ...

device

The device the module is on. Use it to keep your code device agnostic.

def training_step(self, batch, batch_idx):
    z = torch.rand(2, 3, device=self.device)

global_rank

The global_rank is the index of the current process across all nodes and devices. Lightning will perform some operations such as logging, weight checkpointing only when global_rank=0. You usually do not need to use this property, but it is useful to know how to access it if needed.

def training_step(self, batch, batch_idx):
    if self.global_rank == 0:
        # do something only once across all the nodes
        ...

global_step

The number of optimizer steps taken (does not reset each epoch). This includes multiple optimizers and TBPTT steps (if enabled).

def training_step(self, batch, batch_idx):
    self.logger.experiment.log_image(..., step=self.global_step)

hparams

The arguments passed through LightningModule.__init__() and saved by calling save_hyperparameters() could be accessed by the hparams attribute.

def __init__(self, learning_rate):
    self.save_hyperparameters()


def configure_optimizers(self):
    return Adam(self.parameters(), lr=self.hparams.learning_rate)

logger

The current logger being used (tensorboard or other supported logger)

def training_step(self, batch, batch_idx):
    # the generic logger (same no matter if tensorboard or other supported logger)
    self.logger

    # the particular logger
    tensorboard_logger = self.logger.experiment

loggers

The list of loggers currently being used by the Trainer.

def training_step(self, batch, batch_idx):
    # List of Logger objects
    loggers = self.loggers
    for logger in loggers:
        logger.log_metrics({"foo": 1.0})

local_rank

The local_rank is the index of the current process across all the devices for the current node. You usually do not need to use this property, but it is useful to know how to access it if needed. For example, if using 10 machines (or nodes), the GPU at index 0 on each machine has local_rank = 0.

def training_step(self, batch, batch_idx):
    if self.local_rank == 0:
        # do something only once across each node
        ...

precision

The type of precision used:

def training_step(self, batch, batch_idx):
    if self.precision == 16:
        ...

trainer

Pointer to the trainer

def training_step(self, batch, batch_idx):
    max_steps = self.trainer.max_steps
    any_flag = self.trainer.any_flag

prepare_data_per_node

If set to True will call prepare_data() on LOCAL_RANK=0 for every node. If set to False will only call from NODE_RANK=0, LOCAL_RANK=0.

class LitModel(LightningModule):
    def __init__(self):
        super().__init__()
        self.prepare_data_per_node = True

automatic_optimization

When set to False, Lightning does not automate the optimization process. This means you are responsible for handling your optimizers. However, we do take care of precision and any accelerators used.

See manual optimization for details.

def __init__(self):
    self.automatic_optimization = False


def training_step(self, batch, batch_idx):
    opt = self.optimizers(use_pl_optimizer=True)

    loss = ...
    opt.zero_grad()
    self.manual_backward(loss)
    opt.step()

This is recommended only if using 2+ optimizers AND if you know how to perform the optimization procedure properly. Note that automatic optimization can still be used with multiple optimizers by relying on the optimizer_idx parameter. Manual optimization is most useful for research topics like reinforcement learning, sparse coding, and GAN research.

def __init__(self):
    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)

    gen_loss = ...
    opt_a.zero_grad()
    self.manual_backward(gen_loss)
    opt_a.step()

    disc_loss = ...
    opt_b.zero_grad()
    self.manual_backward(disc_loss)
    opt_b.step()

example_input_array

Set and access example_input_array, which basically represents a single batch.

def __init__(self):
    self.example_input_array = ...
    self.generator = ...


def on_train_epoch_end(self):
    # generate some images using the example_input_array
    gen_images = self.generator(self.example_input_array)

truncated_bptt_steps

Truncated Backpropagation Through Time (TBPTT) performs perform backpropogation every k steps of a much longer sequence. This is made possible by passing training batches split along the time-dimensions into splits of size k to the training_step. In order to keep the same forward propagation behavior, all hidden states should be kept in-between each time-dimension split.

If this is enabled, your batches will automatically get truncated and the Trainer will apply Truncated Backprop to it.

(Williams et al. “An efficient gradient-based algorithm for on-line training of recurrent network trajectories.”)

Tutorial

from pytorch_lightning import LightningModule


class MyModel(LightningModule):
    def __init__(self, input_size, hidden_size, num_layers):
        super().__init__()
        # batch_first has to be set to True
        self.lstm = nn.LSTM(
            input_size=input_size,
            hidden_size=hidden_size,
            num_layers=num_layers,
            batch_first=True,
        )

        ...

        # Important: This property activates truncated backpropagation through time
        # Setting this value to 2 splits the batch into sequences of size 2
        self.truncated_bptt_steps = 2

    # Truncated back-propagation through time
    def training_step(self, batch, batch_idx, hiddens):
        x, y = batch

        # the training step must be updated to accept a ``hiddens`` argument
        # hiddens are the hiddens from the previous truncated backprop step
        out, hiddens = self.lstm(x, hiddens)

        ...

        return {"loss": ..., "hiddens": hiddens}

Lightning takes care of splitting your batch along the time-dimension. It is assumed to be the second dimension of your batches. Therefore, in the example above, we have set batch_first=True.

# we use the second as the time dimension
# (batch, time, ...)
sub_batch = batch[0, 0:t, ...]

To modify how the batch is split, override the pytorch_lightning.core.module.LightningModule.tbptt_split_batch() method:

class LitMNIST(LightningModule):
    def tbptt_split_batch(self, batch, split_size):
        # do your own splitting on the batch
        return splits

Hooks

This is the pseudocode to describe the structure of fit(). The inputs and outputs of each function are not represented for simplicity. Please check each function’s API reference for more information.

def fit(self):
    if global_rank == 0:
        # prepare data is called on GLOBAL_ZERO only
        prepare_data()

    configure_callbacks()

    with parallel(devices):
        # devices can be GPUs, TPUs, ...
        train_on_device(model)


def train_on_device(model):
    # called PER DEVICE
    setup("fit")
    configure_optimizers()
    on_fit_start()

    # the sanity check runs here

    on_train_start()
    for epoch in epochs:
        fit_loop()
    on_train_end()

    on_fit_end()
    teardown("fit")


def fit_loop():
    on_train_epoch_start()

    for batch in train_dataloader():
        on_train_batch_start()

        on_before_batch_transfer()
        transfer_batch_to_device()
        on_after_batch_transfer()

        training_step()

        on_before_zero_grad()
        optimizer_zero_grad()

        on_before_backward()
        backward()
        on_after_backward()

        on_before_optimizer_step()
        configure_gradient_clipping()
        optimizer_step()

        on_train_batch_end()

        if should_check_val:
            val_loop()
    # end training epoch
    training_epoch_end()

    on_train_epoch_end()


def val_loop():
    on_validation_model_eval()  # calls `model.eval()`
    torch.set_grad_enabled(False)

    on_validation_start()
    on_validation_epoch_start()

    val_outs = []
    for batch_idx, batch in enumerate(val_dataloader()):
        on_validation_batch_start(batch, batch_idx)

        batch = on_before_batch_transfer(batch)
        batch = transfer_batch_to_device(batch)
        batch = on_after_batch_transfer(batch)

        out = validation_step(batch, batch_idx)

        on_validation_batch_end(batch, batch_idx)
        val_outs.append(out)

    validation_epoch_end(val_outs)

    on_validation_epoch_end()
    on_validation_end()

    # set up for train
    on_validation_model_train()  # calls `model.train()`
    torch.set_grad_enabled(True)

backward

LightningModule.backward(loss, optimizer, optimizer_idx, *args, **kwargs)[source]

Called to perform backward on the loss returned in training_step(). Override this hook with your own implementation if you need to.

Parameters
  • loss (Tensor) – The loss tensor returned by training_step(). If gradient accumulation is used, the loss here holds the normalized value (scaled by 1 / accumulation steps).

  • optimizer (Optional[Steppable]) – Current optimizer being used. None if using manual optimization.

  • optimizer_idx (Optional[int]) – Index of the current optimizer being used. None if using manual optimization.

Example:

def backward(self, loss, optimizer, optimizer_idx):
    loss.backward()
Return type

None

on_before_backward

LightningModule.on_before_backward(loss)

Called before loss.backward().

Parameters

loss (Tensor) – Loss divided by number of batches for gradient accumulation and scaled if using native AMP.

Return type

None

on_after_backward

LightningModule.on_after_backward()

Called after loss.backward() and before optimizers are stepped.

Note

If using native AMP, the gradients will not be unscaled at this point. Use the on_before_optimizer_step if you need the unscaled gradients.

Return type

None

on_before_zero_grad

LightningModule.on_before_zero_grad(optimizer)

Called after training_step() and before optimizer.zero_grad().

Called in the training loop after taking an optimizer step and before zeroing grads. Good place to inspect weight information with weights updated.

This is where it is called:

for optimizer in optimizers:
    out = training_step(...)

    model.on_before_zero_grad(optimizer) # < ---- called here
    optimizer.zero_grad()

    backward()
Parameters

optimizer (Optimizer) – The optimizer for which grads should be zeroed.

Return type

None

on_fit_start

LightningModule.on_fit_start()

Called at the very beginning of fit.

If on DDP it is called on every process

Return type

None

on_fit_end

LightningModule.on_fit_end()

Called at the very end of fit.

If on DDP it is called on every process

Return type

None

on_load_checkpoint

LightningModule.on_load_checkpoint(checkpoint)

Called by Lightning to restore your model. If you saved something with on_save_checkpoint() this is your chance to restore this.

Parameters

checkpoint (Dict[str, Any]) – Loaded checkpoint

Example:

def on_load_checkpoint(self, checkpoint):
    # 99% of the time you don't need to implement this method
    self.something_cool_i_want_to_save = checkpoint['something_cool_i_want_to_save']

Note

Lightning auto-restores global step, epoch, and train state including amp scaling. There is no need for you to restore anything regarding training.

Return type

None

on_save_checkpoint

LightningModule.on_save_checkpoint(checkpoint)

Called by Lightning when saving a checkpoint to give you a chance to store anything else you might want to save.

Parameters

checkpoint (Dict[str, Any]) – The full checkpoint dictionary before it gets dumped to a file. Implementations of this hook can insert additional data into this dictionary.

Example:

def on_save_checkpoint(self, checkpoint):
    # 99% of use cases you don't need to implement this method
    checkpoint['something_cool_i_want_to_save'] = my_cool_pickable_object

Note

Lightning saves all aspects of training (epoch, global step, etc…) including amp scaling. There is no need for you to store anything about training.

Return type

None

load_from_checkpoint

classmethod LightningModule.load_from_checkpoint(checkpoint_path, map_location=None, hparams_file=None, strict=True, **kwargs)

Primary way of loading a model from a checkpoint. When Lightning saves a checkpoint it stores the arguments passed to __init__ in the checkpoint under "hyper_parameters".

Any arguments specified through **kwargs will override args stored in "hyper_parameters".

Parameters
  • checkpoint_path (Union[str, Path, IO]) – Path to checkpoint. This can also be a URL, or file-like object

  • map_location (Union[device, str, int, Callable[[Union[device, str, int]], Union[device, str, int]], Dict[Union[device, str, int], Union[device, str, int]], None]) – If your checkpoint saved a GPU model and you now load on CPUs or a different number of GPUs, use this to map to the new setup. The behaviour is the same as in torch.load().

  • hparams_file (Union[str, Path, None]) –

    Optional path to a .yaml or .csv file with hierarchical structure as in this example:

    drop_prob: 0.2
    dataloader:
        batch_size: 32
    

    You most likely won’t need this since Lightning will always save the hyperparameters to the checkpoint. However, if your checkpoint weights don’t have the hyperparameters saved, use this method to pass in a .yaml file with the hparams you’d like to use. These will be converted into a dict and passed into your LightningModule for use.

    If your model’s hparams argument is Namespace and .yaml file has hierarchical structure, you need to refactor your model to treat hparams as dict.

  • strict (bool) – Whether to strictly enforce that the keys in checkpoint_path match the keys returned by this module’s state dict.

  • **kwargs – Any extra keyword args needed to init the model. Can also be used to override saved hyperparameter values.

Return type

Self

Returns

LightningModule instance with loaded weights and hyperparameters (if available).

Note

load_from_checkpoint is a class method. You should use your LightningModule class to call it instead of the LightningModule instance.

Example:

# load weights without mapping ...
model = MyLightningModule.load_from_checkpoint('path/to/checkpoint.ckpt')

# or load weights mapping all weights from GPU 1 to GPU 0 ...
map_location = {'cuda:1':'cuda:0'}
model = MyLightningModule.load_from_checkpoint(
    'path/to/checkpoint.ckpt',
    map_location=map_location
)

# or load weights and hyperparameters from separate files.
model = MyLightningModule.load_from_checkpoint(
    'path/to/checkpoint.ckpt',
    hparams_file='/path/to/hparams_file.yaml'
)

# override some of the params with new values
model = MyLightningModule.load_from_checkpoint(
    PATH,
    num_layers=128,
    pretrained_ckpt_path=NEW_PATH,
)

# predict
pretrained_model.eval()
pretrained_model.freeze()
y_hat = pretrained_model(x)

on_train_start

LightningModule.on_train_start()

Called at the beginning of training after sanity check.

Return type

None

on_train_end

LightningModule.on_train_end()

Called at the end of training before logger experiment is closed.

Return type

None

on_validation_start

LightningModule.on_validation_start()

Called at the beginning of validation.

Return type

None

on_validation_end

LightningModule.on_validation_end()

Called at the end of validation.

Return type

None

on_test_batch_start

LightningModule.on_test_batch_start(batch, batch_idx, dataloader_idx)

Called in the test loop before anything happens for that batch.

Parameters
  • batch (Any) – The batched data as it is returned by the test DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_test_batch_end

LightningModule.on_test_batch_end(outputs, batch, batch_idx, dataloader_idx)

Called in the test loop after the batch.

Parameters
  • outputs (Union[Tensor, Dict[str, Any], None]) – The outputs of test_step_end(test_step(x))

  • batch (Any) – The batched data as it is returned by the test DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_test_epoch_start

LightningModule.on_test_epoch_start()

Called in the test loop at the very beginning of the epoch.

Return type

None

on_test_epoch_end

LightningModule.on_test_epoch_end()

Called in the test loop at the very end of the epoch.

Return type

None

on_test_start

LightningModule.on_test_start()

Called at the beginning of testing.

Return type

None

on_test_end

LightningModule.on_test_end()

Called at the end of testing.

Return type

None

on_predict_batch_start

LightningModule.on_predict_batch_start(batch, batch_idx, dataloader_idx)

Called in the predict loop before anything happens for that batch.

Parameters
  • batch (Any) – The batched data as it is returned by the test DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_predict_batch_end

LightningModule.on_predict_batch_end(outputs, batch, batch_idx, dataloader_idx)

Called in the predict loop after the batch.

Parameters
  • outputs (Optional[Any]) – The outputs of predict_step_end(test_step(x))

  • batch (Any) – The batched data as it is returned by the test DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_predict_epoch_start

LightningModule.on_predict_epoch_start()

Called at the beginning of predicting.

Return type

None

on_predict_epoch_end

LightningModule.on_predict_epoch_end(results)

Called at the end of predicting.

Return type

None

on_predict_start

LightningModule.on_predict_start()

Called at the beginning of predicting.

Return type

None

on_predict_end

LightningModule.on_predict_end()

Called at the end of predicting.

Return type

None

on_train_batch_start

LightningModule.on_train_batch_start(batch, batch_idx)

Called in the training loop before anything happens for that batch.

If you return -1 here, you will skip training for the rest of the current epoch.

Parameters
  • batch (Any) – The batched data as it is returned by the training DataLoader.

  • batch_idx (int) – the index of the batch

Return type

Optional[int]

on_train_batch_end

LightningModule.on_train_batch_end(outputs, batch, batch_idx)

Called in the training loop after the batch.

Parameters
  • outputs (Union[Tensor, Dict[str, Any]]) – The outputs of training_step_end(training_step(x))

  • batch (Any) – The batched data as it is returned by the training DataLoader.

  • batch_idx (int) – the index of the batch

Return type

None

on_train_epoch_start

LightningModule.on_train_epoch_start()

Called in the training loop at the very beginning of the epoch.

Return type

None

on_train_epoch_end

LightningModule.on_train_epoch_end()

Called in the training loop at the very end of the epoch.

To access all batch outputs at the end of the epoch, either:

  1. Implement training_epoch_end in the LightningModule OR

  2. Cache data across steps on the attribute(s) of the LightningModule and access them in this hook

Return type

None

on_validation_batch_start

LightningModule.on_validation_batch_start(batch, batch_idx, dataloader_idx)

Called in the validation loop before anything happens for that batch.

Parameters
  • batch (Any) – The batched data as it is returned by the validation DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_validation_batch_end

LightningModule.on_validation_batch_end(outputs, batch, batch_idx, dataloader_idx)

Called in the validation loop after the batch.

Parameters
  • outputs (Union[Tensor, Dict[str, Any], None]) – The outputs of validation_step_end(validation_step(x))

  • batch (Any) – The batched data as it is returned by the validation DataLoader.

  • batch_idx (int) – the index of the batch

  • dataloader_idx (int) – the index of the dataloader

Return type

None

on_validation_epoch_start

LightningModule.on_validation_epoch_start()

Called in the validation loop at the very beginning of the epoch.

Return type

None

on_validation_epoch_end

LightningModule.on_validation_epoch_end()

Called in the validation loop at the very end of the epoch.

Return type

None

configure_sharded_model

LightningModule.configure_sharded_model()

Hook to create modules in a distributed aware context. This is useful for when using sharded plugins, where we’d like to shard the model instantly, which is useful for extremely large models which can save memory and initialization time.

This hook is called during each of fit/val/test/predict stages in the same process, so ensure that implementation of this hook is idempotent.

Return type

None

on_validation_model_eval

LightningModule.on_validation_model_eval()

Sets the model to eval during the val loop.

Return type

None

on_validation_model_train

LightningModule.on_validation_model_train()

Sets the model to train during the val loop.

Return type

None

on_test_model_eval

LightningModule.on_test_model_eval()

Sets the model to eval during the test loop.

Return type

None

on_test_model_train

LightningModule.on_test_model_train()

Sets the model to train during the test loop.

Return type

None

on_before_optimizer_step

LightningModule.on_before_optimizer_step(optimizer, optimizer_idx)

Called before optimizer.step().

If using gradient accumulation, the hook is called once the gradients have been accumulated. See: accumulate_grad_batches.

If using native AMP, the loss will be unscaled before calling this hook. See these docs for more information on the scaling of gradients.

If clipping gradients, the gradients will not have been clipped yet.

Parameters
  • optimizer (Optimizer) – Current optimizer being used.

  • optimizer_idx (int) – Index of the current optimizer being used.

Example:

def on_before_optimizer_step(self, optimizer, optimizer_idx):
    # example to inspect gradient information in tensorboard
    if self.trainer.global_step % 25 == 0:  # don't make the tf file huge
        for k, v in self.named_parameters():
            self.logger.experiment.add_histogram(
                tag=k, values=v.grad, global_step=self.trainer.global_step
            )
Return type

None

configure_gradient_clipping

LightningModule.configure_gradient_clipping(optimizer, optimizer_idx, gradient_clip_val=None, gradient_clip_algorithm=None)[source]

Perform gradient clipping for the optimizer parameters. Called before optimizer_step().

Parameters
  • optimizer (Optimizer) – Current optimizer being used.

  • optimizer_idx (int) – Index of the current optimizer being used.

  • gradient_clip_val (Union[int, float, None]) – The value at which to clip gradients. By default value passed in Trainer will be available here.

  • gradient_clip_algorithm (Optional[str]) – The gradient clipping algorithm to use. By default value passed in Trainer will be available here.

Example:

# Perform gradient clipping on gradients associated with discriminator (optimizer_idx=1) in GAN
def configure_gradient_clipping(self, optimizer, optimizer_idx, gradient_clip_val, gradient_clip_algorithm):
    if optimizer_idx == 1:
        # Lightning will handle the gradient clipping
        self.clip_gradients(
            optimizer,
            gradient_clip_val=gradient_clip_val,
            gradient_clip_algorithm=gradient_clip_algorithm
        )
    else:
        # implement your own custom logic to clip gradients for generator (optimizer_idx=0)
Return type

None

optimizer_step

LightningModule.optimizer_step(epoch, batch_idx, optimizer, optimizer_idx=0, optimizer_closure=None, on_tpu=False, using_lbfgs=False)[source]

Override this method to adjust the default way the Trainer calls each optimizer.

By default, Lightning calls step() and zero_grad() as shown in the example once per optimizer. This method (and zero_grad()) won’t be called during the accumulation phase when Trainer(accumulate_grad_batches != 1). Overriding this hook has no benefit with manual optimization.

Parameters
  • epoch (int) – Current epoch

  • batch_idx (int) – Index of current batch

  • optimizer (Union[Optimizer, LightningOptimizer]) – A PyTorch optimizer

  • optimizer_idx (int) – If you used multiple optimizers, this indexes into that list.

  • optimizer_closure (Optional[Callable[[], Any]]) – The optimizer closure. This closure must be executed as it includes the calls to training_step(), optimizer.zero_grad(), and backward().

  • on_tpu (bool) – True if TPU backward is required

  • using_lbfgs (bool) – True if the matching optimizer is torch.optim.LBFGS

Examples:

# DEFAULT
def optimizer_step(self, epoch, batch_idx, optimizer, optimizer_idx,
                   optimizer_closure, on_tpu, using_lbfgs):
    optimizer.step(closure=optimizer_closure)

# Alternating schedule for optimizer steps (i.e.: GANs)
def optimizer_step(self, epoch, batch_idx, optimizer, optimizer_idx,
                   optimizer_closure, on_tpu, using_lbfgs):
    # update generator opt every step
    if optimizer_idx == 0:
        optimizer.step(closure=optimizer_closure)

    # update discriminator opt every 2 steps
    if optimizer_idx == 1:
        if (batch_idx + 1) % 2 == 0 :
            optimizer.step(closure=optimizer_closure)
        else:
            # call the closure by itself to run `training_step` + `backward` without an optimizer step
            optimizer_closure()

    # ...
    # add as many optimizers as you want

Here’s another example showing how to use this for more advanced things such as learning rate warm-up:

# learning rate warm-up
def optimizer_step(
    self,
    epoch,
    batch_idx,
    optimizer,
    optimizer_idx,
    optimizer_closure,
    on_tpu,
    using_lbfgs,
):
    # update params
    optimizer.step(closure=optimizer_closure)

    # manually warm up lr without a scheduler
    if self.trainer.global_step < 500:
        lr_scale = min(1.0, float(self.trainer.global_step + 1) / 500.0)
        for pg in optimizer.param_groups:
            pg["lr"] = lr_scale * self.learning_rate
Return type

None

optimizer_zero_grad

LightningModule.optimizer_zero_grad(epoch, batch_idx, optimizer, optimizer_idx)[source]

Override this method to change the default behaviour of optimizer.zero_grad().

Parameters
  • epoch (int) – Current epoch

  • batch_idx (int) – Index of current batch

  • optimizer (Optimizer) – A PyTorch optimizer

  • optimizer_idx (int) – If you used multiple optimizers this indexes into that list.

Examples:

# DEFAULT
def optimizer_zero_grad(self, epoch, batch_idx, optimizer, optimizer_idx):
    optimizer.zero_grad()

# Set gradients to `None` instead of zero to improve performance (not required on `torch>=2.0.0`).
def optimizer_zero_grad(self, epoch, batch_idx, optimizer, optimizer_idx):
    optimizer.zero_grad(set_to_none=True)

See torch.optim.Optimizer.zero_grad() for the explanation of the above example.

Return type

None

prepare_data

LightningModule.prepare_data()

Use this to download and prepare data. Downloading and saving data with multiple processes (distributed settings) will result in corrupted data. Lightning ensures this method is called only within a single process, so you can safely add your downloading logic within.

Warning

DO NOT set state to the model (use setup instead) since this is NOT called on every device

Example:

def prepare_data(self):
    # good
    download_data()
    tokenize()
    etc()

    # bad
    self.split = data_split
    self.some_state = some_other_state()

In a distributed environment, prepare_data can be called in two ways (using prepare_data_per_node)

  1. Once per node. This is the default and is only called on LOCAL_RANK=0.

  2. Once in total. Only called on GLOBAL_RANK=0.

Example:

# DEFAULT
# called once per node on LOCAL_RANK=0 of that node
class LitDataModule(LightningDataModule):
    def __init__(self):
        super().__init__()
        self.prepare_data_per_node = True


# call on GLOBAL_RANK=0 (great for shared file systems)
class LitDataModule(LightningDataModule):
    def __init__(self):
        super().__init__()
        self.prepare_data_per_node = False

This is called before requesting the dataloaders:

model.prepare_data()
initialize_distributed()
model.setup(stage)
model.train_dataloader()
model.val_dataloader()
model.test_dataloader()
model.predict_dataloader()
Return type

None

setup

LightningModule.setup(stage)

Called at the beginning of fit (train + validate), validate, test, or predict. This is a good hook when you need to build models dynamically or adjust something about them. This hook is called on every process when using DDP.

Parameters

stage (str) – either 'fit', 'validate', 'test', or 'predict'

Example:

class LitModel(...):
    def __init__(self):
        self.l1 = None

    def prepare_data(self):
        download_data()
        tokenize()

        # don't do this
        self.something = else

    def setup(self, stage):
        data = load_data(...)
        self.l1 = nn.Linear(28, data.num_classes)
Return type

None

tbptt_split_batch

LightningModule.tbptt_split_batch(batch, split_size)[source]

When using truncated backpropagation through time, each batch must be split along the time dimension. Lightning handles this by default, but for custom behavior override this function.

Parameters
  • batch (Any) – Current batch

  • split_size (int) – The size of the split

Return type

List[Any]

Returns

List of batch splits. Each split will be passed to training_step() to enable truncated back propagation through time. The default implementation splits root level Tensors and Sequences at dim=1 (i.e. time dim). It assumes that each time dim is the same length.

Examples:

def tbptt_split_batch(self, batch, split_size):
    splits = []
    for t in range(0, time_dims[0], split_size):
        batch_split = []
        for i, x in enumerate(batch):
            if isinstance(x, torch.Tensor):
                split_x = x[:, t:t + split_size]
            elif isinstance(x, collections.abc.Sequence):
                split_x = [None] * len(x)
                for batch_idx in range(len(x)):
                  split_x[batch_idx] = x[batch_idx][t:t + split_size]
            batch_split.append(split_x)
        splits.append(batch_split)
    return splits

Note

Called in the training loop after on_train_batch_start() if truncated_bptt_steps > 0. Each returned batch split is passed separately to training_step().

teardown

LightningModule.teardown(stage)

Called at the end of fit (train + validate), validate, test, or predict.

Parameters

stage (str) – either 'fit', 'validate', 'test', or 'predict'

Return type

None

train_dataloader

LightningModule.train_dataloader()

Implement one or more PyTorch DataLoaders for training.

Return type

Union[DataLoader, Sequence[DataLoader], Sequence[Sequence[DataLoader]], Sequence[Dict[str, DataLoader]], Dict[str, DataLoader], Dict[str, Dict[str, DataLoader]], Dict[str, Sequence[DataLoader]]]

Returns

A collection of torch.utils.data.DataLoader specifying training samples. In the case of multiple dataloaders, please see this section.

The dataloader you return will not be reloaded unless you set reload_dataloaders_every_n_epochs to a positive integer.

For data processing use the following pattern:

  • download in prepare_data()

  • process and split in setup()

However, the above are only necessary for distributed processing.

Warning

do not assign state in prepare_data

  • fit()

  • prepare_data()

  • setup()

Note

Lightning adds the correct sampler for distributed and arbitrary hardware. There is no need to set it yourself.

Example:

# single dataloader
def train_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=True, transform=transform,
                    download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.batch_size,
        shuffle=True
    )
    return loader

# multiple dataloaders, return as list
def train_dataloader(self):
    mnist = MNIST(...)
    cifar = CIFAR(...)
    mnist_loader = torch.utils.data.DataLoader(
        dataset=mnist, batch_size=self.batch_size, shuffle=True
    )
    cifar_loader = torch.utils.data.DataLoader(
        dataset=cifar, batch_size=self.batch_size, shuffle=True
    )
    # each batch will be a list of tensors: [batch_mnist, batch_cifar]
    return [mnist_loader, cifar_loader]

# multiple dataloader, return as dict
def train_dataloader(self):
    mnist = MNIST(...)
    cifar = CIFAR(...)
    mnist_loader = torch.utils.data.DataLoader(
        dataset=mnist, batch_size=self.batch_size, shuffle=True
    )
    cifar_loader = torch.utils.data.DataLoader(
        dataset=cifar, batch_size=self.batch_size, shuffle=True
    )
    # each batch will be a dict of tensors: {'mnist': batch_mnist, 'cifar': batch_cifar}
    return {'mnist': mnist_loader, 'cifar': cifar_loader}

val_dataloader

LightningModule.val_dataloader()

Implement one or multiple PyTorch DataLoaders for validation.

The dataloader you return will not be reloaded unless you set reload_dataloaders_every_n_epochs to a positive integer.

It’s recommended that all data downloads and preparation happen in prepare_data().

Note

Lightning adds the correct sampler for distributed and arbitrary hardware There is no need to set it yourself.

Return type

Union[DataLoader, Sequence[DataLoader]]

Returns

A torch.utils.data.DataLoader or a sequence of them specifying validation samples.

Examples:

def val_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=False,
                    transform=transform, download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.batch_size,
        shuffle=False
    )

    return loader

# can also return multiple dataloaders
def val_dataloader(self):
    return [loader_a, loader_b, ..., loader_n]

Note

If you don’t need a validation dataset and a validation_step(), you don’t need to implement this method.

Note

In the case where you return multiple validation dataloaders, the validation_step() will have an argument dataloader_idx which matches the order here.

test_dataloader

LightningModule.test_dataloader()

Implement one or multiple PyTorch DataLoaders for testing.

For data processing use the following pattern:

  • download in prepare_data()

  • process and split in setup()

However, the above are only necessary for distributed processing.

Warning

do not assign state in prepare_data

  • test()

  • prepare_data()

  • setup()

Note

Lightning adds the correct sampler for distributed and arbitrary hardware. There is no need to set it yourself.

Return type

Union[DataLoader, Sequence[DataLoader]]

Returns

A torch.utils.data.DataLoader or a sequence of them specifying testing samples.

Example:

def test_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=False, transform=transform,
                    download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.batch_size,
        shuffle=False
    )

    return loader

# can also return multiple dataloaders
def test_dataloader(self):
    return [loader_a, loader_b, ..., loader_n]

Note

If you don’t need a test dataset and a test_step(), you don’t need to implement this method.

Note

In the case where you return multiple test dataloaders, the test_step() will have an argument dataloader_idx which matches the order here.

predict_dataloader

LightningModule.predict_dataloader()

Implement one or multiple PyTorch DataLoaders for prediction.

It’s recommended that all data downloads and preparation happen in prepare_data().

Note

Lightning adds the correct sampler for distributed and arbitrary hardware There is no need to set it yourself.

Return type

Union[DataLoader, Sequence[DataLoader]]

Returns

A torch.utils.data.DataLoader or a sequence of them specifying prediction samples.

Note

In the case where you return multiple prediction dataloaders, the predict_step() will have an argument dataloader_idx which matches the order here.

transfer_batch_to_device

LightningModule.transfer_batch_to_device(batch, device, dataloader_idx)

Override this hook if your DataLoader returns tensors wrapped in a custom data structure.

The data types listed below (and any arbitrary nesting of them) are supported out of the box:

For anything else, you need to define how the data is moved to the target device (CPU, GPU, TPU, …).

Note

This hook should only transfer the data and not modify it, nor should it move the data to any other device than the one passed in as argument (unless you know what you are doing). To check the current state of execution of this hook you can use self.trainer.training/testing/validating/predicting so that you can add different logic as per your requirement.

Note

This hook only runs on single GPU training and DDP (no data-parallel). Data-Parallel support will come in near future.

Parameters
  • batch (Any) – A batch of data that needs to be transferred to a new device.

  • device (device) – The target device as defined in PyTorch.

  • dataloader_idx (int) – The index of the dataloader to which the batch belongs.

Return type

Any

Returns

A reference to the data on the new device.

Example:

def transfer_batch_to_device(self, batch, device, dataloader_idx):
    if isinstance(batch, CustomBatch):
        # move all tensors in your custom data structure to the device
        batch.samples = batch.samples.to(device)
        batch.targets = batch.targets.to(device)
    elif dataloader_idx == 0:
        # skip device transfer for the first dataloader or anything you wish
        pass
    else:
        batch = super().transfer_batch_to_device(batch, device, dataloader_idx)
    return batch
Raises
  • MisconfigurationException – If using data-parallel, Trainer(strategy='dp').

  • MisconfigurationException – If using IPUs, Trainer(accelerator='ipu').

See also

  • move_data_to_device()

  • apply_to_collection()

on_before_batch_transfer

LightningModule.on_before_batch_transfer(batch, dataloader_idx)

Override to alter or apply batch augmentations to your batch before it is transferred to the device.

Note

To check the current state of execution of this hook you can use self.trainer.training/testing/validating/predicting so that you can add different logic as per your requirement.

Note

This hook only runs on single GPU training and DDP (no data-parallel). Data-Parallel support will come in near future.

Parameters
  • batch (Any) – A batch of data that needs to be altered or augmented.

  • dataloader_idx (int) – The index of the dataloader to which the batch belongs.

Return type

Any

Returns

A batch of data

Example:

def on_before_batch_transfer(self, batch, dataloader_idx):
    batch['x'] = transforms(batch['x'])
    return batch

See also

  • on_after_batch_transfer()

  • transfer_batch_to_device()

on_after_batch_transfer

LightningModule.on_after_batch_transfer(batch, dataloader_idx)

Override to alter or apply batch augmentations to your batch after it is transferred to the device.

Note

To check the current state of execution of this hook you can use self.trainer.training/testing/validating/predicting so that you can add different logic as per your requirement.

Note

This hook only runs on single GPU training and DDP (no data-parallel). Data-Parallel support will come in near future.

Parameters
  • batch (Any) – A batch of data that needs to be altered or augmented.

  • dataloader_idx (int) – The index of the dataloader to which the batch belongs.

Return type

Any

Returns

A batch of data

Example:

def on_after_batch_transfer(self, batch, dataloader_idx):
    batch['x'] = gpu_transforms(batch['x'])
    return batch
Raises
  • MisconfigurationException – If using data-parallel, Trainer(strategy='dp').

  • MisconfigurationException – If using IPUs, Trainer(accelerator='ipu').

See also

  • on_before_batch_transfer()

  • transfer_batch_to_device()