lightning¶

Classes

LightningModule

nn.Module with additional great features.

class pytorch_lightning.core.lightning.LightningModule(*args, **kwargs)[source]¶

Bases: abc.ABC, pytorch_lightning.utilities.device_dtype_mixin.DeviceDtypeModuleMixin, pytorch_lightning.core.grads.GradInformation, pytorch_lightning.core.saving.ModelIO, pytorch_lightning.core.hooks.ModelHooks, pytorch_lightning.core.hooks.DataHooks, pytorch_lightning.core.hooks.CheckpointHooks, torch.nn.Module

all_gather(tensor, 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

tensor¶ (Tensor) – tensor of shape (batch, …)
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 all_gather op

Returns

A tensor of shape (world_size, batch, …)

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

Override backward with your own implementation if you need to.

Parameters

loss¶ (Tensor) – Loss is already scaled by accumulated grads
optimizer¶ (Optimizer) – Current optimizer being used
optimizer_idx¶ (int) – Index of the current optimizer being used

Called to perform backward step. Feel free to override as needed. The loss passed in has already been scaled for accumulated gradients if requested.

Example:

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

Return type: None

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.

Returns

Any of these 6 options.

Single optimizer.
List or Tuple - List of optimizers.
Two lists - The first list has multiple optimizers, the second a list of LR schedulers (or lr_dict).
Dictionary, with an ‘optimizer’ key, and (optionally) a ‘lr_scheduler’ key whose value is a single LR scheduler or lr_dict.
Tuple of dictionaries as described, with an optional ‘frequency’ key.
None - Fit will run without any optimizer.

Note

The ‘frequency’ value 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.

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

{
    'scheduler': lr_scheduler, # The LR scheduler instance (required)
    'interval': 'epoch', # The unit of the scheduler's step size
    'frequency': 1, # The frequency of the scheduler
    'reduce_on_plateau': False, # For ReduceLROnPlateau scheduler
    'monitor': 'val_loss', # Metric for ReduceLROnPlateau to monitor
    'strict': True, # Whether to crash the training if `monitor` is not found
    'name': None, # Custom name for LearningRateMonitor to use
}

Only the scheduler key is required, the rest will be set to the defaults above.

Examples

# most cases
def configure_optimizers(self):
    opt = Adam(self.parameters(), lr=1e-3)
    return opt

# multiple optimizer case (e.g.: GAN)
def configure_optimizers(self):
    generator_opt = Adam(self.model_gen.parameters(), lr=0.01)
    disriminator_opt = Adam(self.model_disc.parameters(), lr=0.02)
    return generator_opt, disriminator_opt

# example with learning rate schedulers
def configure_optimizers(self):
    generator_opt = Adam(self.model_gen.parameters(), lr=0.01)
    disriminator_opt = Adam(self.model_disc.parameters(), lr=0.02)
    discriminator_sched = CosineAnnealing(discriminator_opt, T_max=10)
    return [generator_opt, disriminator_opt], [discriminator_sched]

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

# 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_disc.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 and learning rate scheduler as needed.
If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizers for you.
If you use multiple optimizers, training_step() will have an additional optimizer_idx parameter.
If you use 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.

If you only want to call a learning rate scheduler every x step or epoch, or want to monitor a custom metric, you can specify these in a lr_dict:

{
    'scheduler': lr_scheduler,
    'interval': 'step',  # or 'epoch'
    'monitor': 'val_f1',
    'frequency': x,
}

forward(*args, **kwargs)[source]¶

Same as torch.nn.Module.forward(), however in Lightning you want this to define the operations you want to use for prediction (i.e.: on a server or as a feature extractor).

Normally you’d call self() from your training_step() method. This makes it easy to write a complex system for training with the outputs you’d want in a prediction setting.

You may also find the auto_move_data() decorator useful when using the module outside Lightning in a production setting.

Parameters

*args¶ – Whatever you decide to pass into the forward method.
**kwargs¶ – Keyword arguments are also possible.

Returns

Predicted output

Examples

# example if we were using this model as a feature extractor
def forward(self, x):
    feature_maps = self.convnet(x)
    return feature_maps

def training_step(self, batch, batch_idx):
    x, y = batch
    feature_maps = self(x)
    logits = self.classifier(feature_maps)

    # ...
    return loss

# splitting it this way allows model to be used a feature extractor
model = MyModelAbove()

inputs = server.get_request()
results = model(inputs)
server.write_results(results)

# -------------
# This is in stark contrast to torch.nn.Module where normally you would have this:
def forward(self, batch):
    x, y = batch
    feature_maps = self.convnet(x)
    logits = self.classifier(feature_maps)
    return logits

freeze()[source]¶

Freeze all params for inference.

Example

model = MyLightningModule(...)
model.freeze()

Return type: None

get_progress_bar_dict()[source]¶

Implement this to override the default items displayed in the progress bar. By default it includes the average loss value, split index of BPTT (if used) and the version of the experiment when using a logger.

Epoch 1:   4%|▎         | 40/1095 [00:03<01:37, 10.84it/s, loss=4.501, v_num=10]

Here is an example how to override the defaults:

def get_progress_bar_dict(self):
    # don't show the version number
    items = super().get_progress_bar_dict()
    items.pop("v_num", None)
    return items

Return type: Dict[str, Union[int, str]]
Returns: Dictionary with the items to be displayed in the progress bar.

log(name, value, prog_bar=False, logger=True, on_step=None, on_epoch=None, reduce_fx=torch.mean, tbptt_reduce_fx=torch.mean, tbptt_pad_token=0, enable_graph=False, sync_dist=False, sync_dist_op='mean', sync_dist_group=None)[source]¶

Log a key, value

Example:

self.log('train_loss', loss)

The default behavior per hook is as follows

`*` also applies to the test loop¶
LightningMoule Hook	on_step	on_epoch	prog_bar	logger
training_step	T	F	F	T
training_step_end	T	F	F	T
training_epoch_end	F	T	F	T
validation_step*	F	T	F	T
validation_step_end*	F	T	F	T
validation_epoch_end*	F	T	F	T

Parameters

name¶ (str) – key name
value¶ (Any) – value name
prog_bar¶ (bool) – if True logs to the progress bar
logger¶ (bool) – if True logs to the logger
on_step¶ (Optional[bool]) – if True logs at this step. None auto-logs at the training_step but not validation/test_step
on_epoch¶ (Optional[bool]) – if True logs epoch accumulated metrics. None auto-logs at the val/test step but not training_step
reduce_fx¶ (Callable) – reduction function over step values for end of epoch. Torch.mean by default
tbptt_reduce_fx¶ (Callable) – function to reduce on truncated back prop
tbptt_pad_token¶ (int) – token to use for padding
enable_graph¶ (bool) – if True, will not auto detach the graph
sync_dist¶ (bool) – if True, reduces the metric across GPUs/TPUs
sync_dist_op¶ (Union[Any, str]) – the op to sync across GPUs/TPUs
sync_dist_group¶ (Optional[Any]) – the ddp group

log_dict(dictionary, prog_bar=False, logger=True, on_step=None, on_epoch=None, reduce_fx=torch.mean, tbptt_reduce_fx=torch.mean, tbptt_pad_token=0, enable_graph=False, sync_dist=False, sync_dist_op='mean', sync_dist_group=None)[source]¶

Log a dictonary of values at once

Example:

values = {'loss': loss, 'acc': acc, ..., 'metric_n': metric_n}
self.log_dict(values)

Parameters

dictionary¶ (dict) – key value pairs (str, tensors)
prog_bar¶ (bool) – if True logs to the progress base
logger¶ (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
on_epoch¶ (Optional[bool]) – if True logs epoch accumulated metrics. None auto-logs for val/test step but not training_step
reduce_fx¶ (Callable) – reduction function over step values for end of epoch. Torch.mean by default
tbptt_reduce_fx¶ (Callable) – function to reduce on truncated back prop
tbptt_pad_token¶ (int) – token to use for padding
enable_graph¶ (bool) – if True, will not auto detach the graph
sync_dist¶ (bool) – if True, reduces the metric across GPUs/TPUs
sync_dist_op¶ (Union[Any, str]) – the op to sync across GPUs/TPUs
sync_dist_group¶ (Optional[Any]) – the ddp group:

manual_backward(loss, optimizer, *args, **kwargs)[source]¶

Call this directly from your training_step when doing optimizations manually. By using this we can ensure that all the proper scaling when using 16-bit etc has been done for you

This function forwards all args to the .backward() call as well.

Tip

In manual mode we still automatically clip grads if Trainer(gradient_clip_val=x) is set

Tip

In manual mode we still automatically accumulate grad over batches if Trainer(accumulate_grad_batches=x) is set and you use optimizer.step()

Example:

def training_step(...):
    (opt_a, opt_b) = self.optimizers()
    loss = ...
    # automatically applies scaling, etc...
    self.manual_backward(loss, opt_a)
    opt_a.step()

Return type: None

optimizer_step(epoch=None, batch_idx=None, optimizer=None, optimizer_idx=None, optimizer_closure=None, on_tpu=None, using_native_amp=None, using_lbfgs=None)[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.

Tip

With Trainer(enable_pl_optimizer=True), you can user optimizer.step() directly and it will handle zero_grad, accumulated gradients, AMP, TPU and more automatically for you.

Warning

If you are overriding this method, make sure that you pass the optimizer_closure parameter to optimizer.step() function as shown in the examples. This ensures that train_step_and_backward_closure is called within run_training_batch().

Parameters

epoch¶ (Optional[int]) – Current epoch
batch_idx¶ (Optional[int]) – Index of current batch
optimizer¶ (Optional[Optimizer]) – A PyTorch optimizer
optimizer_idx¶ (Optional[int]) – If you used multiple optimizers this indexes into that list.
optimizer_closure¶ (Optional[Callable]) – closure for all optimizers
on_tpu¶ (Optional[bool]) – true if TPU backward is required
using_native_amp¶ (Optional[bool]) – True if using native amp
using_lbfgs¶ (Optional[bool]) – True if the matching optimizer is lbfgs

Examples

# DEFAULT
def optimizer_step(self, epoch, batch_idx, optimizer, optimizer_idx,
                   optimizer_closure, on_tpu, using_native_amp, 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_native_amp, using_lbfgs):
    # update generator opt every 2 steps
    if optimizer_idx == 0:
        if batch_idx % 2 == 0 :
            optimizer.step(closure=optimizer_closure)
            optimizer.zero_grad()

    # update discriminator opt every 4 steps
    if optimizer_idx == 1:
        if batch_idx % 4 == 0 :
            optimizer.step(closure=optimizer_closure)
            optimizer.zero_grad()

    # ...
    # 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_native_amp, using_lbfgs):
    # warm up lr
    if self.trainer.global_step < 500:
        lr_scale = min(1., float(self.trainer.global_step + 1) / 500.)
        for pg in optimizer.param_groups:
            pg['lr'] = lr_scale * self.learning_rate

    # update params
    optimizer.step(closure=optimizer_closure)
    optimizer.zero_grad()

Return type: None

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. Will be passed to Python’s built-in print function.
**kwargs¶ – Will be passed to Python’s built-in print function.

Example

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

Return type: None

save_hyperparameters(*args, frame=None)[source]¶

Save all model arguments.

Parameters: args¶ – single object of dict, NameSpace or OmegaConf or string names or arguments from class __init__

>>> from collections import OrderedDict
>>> class ManuallyArgsModel(LightningModule):
...     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

>>> class AutomaticArgsModel(LightningModule):
...     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

>>> class SingleArgModel(LightningModule):
...     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

Return type: None

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¶ (Tensor) – Current batch
split_size¶ (int) – The size of the split

Return type

list

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.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_batch_start() if truncated_bptt_steps > 0. Each returned batch split is passed separately to training_step().

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¶ (List[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)

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¶ (Tensor | (Tensor, …) | [Tensor, …]) – The output of your DataLoader. A tensor, tuple or list.
batch_idx¶ (int) – The index of this batch.
dataloader_idx¶ (int) – The index of the dataloader that produced this batch (only if multiple test dataloaders used).

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)

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.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx):
    # 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(*args, **kwargs)[source]¶

Use this when testing with dp or ddp2 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)
batch_parts_outputs = [test_step(sub_batch) for sub_batch in sub_batches]
test_step_end(batch_parts_outputs)

Parameters: batch_parts_outputs¶ – What you return in test_step() for each batch part.
Returns: None or anything

# WITHOUT test_step_end
# if used in DP or DDP2, 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.

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)))

>>> with tempfile.NamedTemporaryFile(suffix='.onnx', delete=False) as tmpfile:
...     model = SimpleModel()
...     input_sample = torch.randn((1, 64))
...     model.to_onnx(tmpfile.name, input_sample, export_params=True)
...     os.path.isfile(tmpfile.name)
True

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 self.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 (Use self.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()
>>> torch.jit.save(model.to_torchscript(), "model.pt")  
>>> os.path.isfile("model.pt")  
>>> torch.jit.save(model.to_torchscript(file_path="model_trace.pt", method='trace', 
...                                     example_inputs=torch.randn(1, 64)))  
>>> os.path.isfile("model_trace.pt")  
True

Return type: Union[ScriptModule, Dict[str, ScriptModule]]
Returns: This LightningModule as a torchscript, regardless of whether file_path is defined or not.

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.

Note

Only called when using multiple optimizers

Override for your own behavior

It works with untoggle_optimizer to make sure param_requires_grad_state is properly reset.

Parameters

optimizer¶ (Optimizer) – Current optimizer used in training_loop
optimizer_idx¶ (int) – Current optimizer idx in training_loop

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 for every 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[Any]) – List of outputs you defined in training_step(), or if there are multiple dataloaders, a list containing a list of outputs for each dataloader.
Return type: None
Returns: None

Note

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

Example:

def training_epoch_end(self, training_step_outputs):
    # do something with all training_step outputs
    return 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 training step for that dataloader.

def training_epoch_end(self, training_step_outputs):
    for out in training_step_outputs:
        # do something here

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¶ (Tensor) – Passed in if truncated_bptt_steps > 0.

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

Note

Returning None is currently not supported for multi-GPU or TPU, or with 16-bit precision enabled.

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)
    ...
    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.

training_step_end(*args, **kwargs)[source]¶

Use this when training with dp or ddp2 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)
batch_parts_outputs = [training_step(sub_batch) for sub_batch in sub_batches]
training_step_end(batch_parts_outputs)

Parameters: batch_parts_outputs¶ – What you return in training_step for each batch part.
Returns: Anything

When using dp/ddp2 distributed backends, 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 denomintaor
    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