trainer

allennlp.training.trainer

Trainer#

class Trainer(Registrable):
 | def __init__(
 |     self,
 |     serialization_dir: str = None,
 |     cuda_device: Optional[Union[int, torch.device]] = None,
 |     distributed: bool = False,
 |     local_rank: int = 0,
 |     world_size: int = 1
 | ) -> None

The base class for an AllenNLP trainer. It can do pretty much anything you want. Your subclass should implement train and also probably from_params.

default_implementation#

class Trainer(Registrable):
 | ...
 | default_implementation = "gradient_descent"

train#

class Trainer(Registrable):
 | ...
 | def train(self) -> Dict[str, Any]

Train a model and return the results.

get_checkpoint_state#

class Trainer(Registrable):
 | ...
 | @contextmanager
 | def get_checkpoint_state(
 |     self
 | ) -> Iterator[Tuple[Dict[str, Any], Dict[str, Any]]]

Returns a tuple of (model state, training state), where training state could have several internal components (e.g., for an, optimizer, learning rate scheduler, etc.).

This is a context manager, and should be called as with trainer.get_checkpoint_state() as state:, so that the trainer has the opportunity to change and restore its internal state for checkpointing. This is used, e.g., for moving averages of model weights.

BatchCallback#

class BatchCallback(Registrable)

An optional callback that you can pass to the GradientDescentTrainer that will be called at the end of every batch, during both training and validation. The default implementation does nothing. You can implement your own callback and do whatever you want, such as saving predictions to disk or extra logging.

call#

class BatchCallback(Registrable):
 | ...
 | def __call__(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     batch_inputs: List[List[TensorDict]],
 |     batch_outputs: List[Dict[str, Any]],
 |     batch_metrics: Dict[str, Any],
 |     epoch: int,
 |     batch_number: int,
 |     is_training: bool,
 |     is_master: bool
 | ) -> None

TensoboardBatchMemoryUsage#

@BatchCallback.register("tensorboard-memory-usage")
class TensoboardBatchMemoryUsage(BatchCallback)

Logs the CPU and GPU memory usage to tensorboard on every batch.

This is mainly used for debugging as it can cause a significant slowdown in training.

call#

class TensoboardBatchMemoryUsage(BatchCallback):
 | ...
 | def __call__(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     batch_inputs: List[List[TensorDict]],
 |     batch_outputs: List[Dict[str, Any]],
 |     batch_metrics: Dict[str, Any],
 |     epoch: int,
 |     batch_number: int,
 |     is_training: bool,
 |     is_master: bool
 | ) -> None

In the distributed case we need to call this from every worker, since every worker reports its own memory usage.

EpochCallback#

class EpochCallback(Registrable)

An optional callback that you can pass to the GradientDescentTrainer that will be called at the end of every epoch (and before the start of training, with epoch=-1). The default implementation does nothing. You can implement your own callback and do whatever you want, such as additional modifications of the trainer's state in between epochs.

call#

class EpochCallback(Registrable):
 | ...
 | def __call__(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     metrics: Dict[str, Any],
 |     epoch: int,
 |     is_master: bool
 | ) -> None

TrackEpochCallback#

@EpochCallback.register("track_epoch_callback")
class TrackEpochCallback:
 | def __init__(self)

A callback that you can pass to the GradientDescentTrainer to access the current epoch number in your model during training. This callback sets model.epoch, which can be read inside of model.forward(). Since the EpochCallback passes epoch=-1 at the start of the training, we set model.epoch = epoch + 1 which now denotes the number of completed epochs at a given training state.

call#

class TrackEpochCallback:
 | ...
 | def __call__(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     metrics: Dict[str, Any],
 |     epoch: int,
 |     is_master: bool
 | ) -> None

TrainerCallback#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta)

A general callback object that wraps all three types of callbacks into one.

Rather than a __call__ method, this class has on_batch, on_epoch, and on_end methods, corresponding to each callback type. Each one receives the state of the wrapper object as self. This enables easier state sharing between related callbacks.

Under the hood, this is a metaclass that creates wrapping subclasses each time a subclass is created.

on_batch#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def on_batch(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     batch_inputs: List[List[TensorDict]],
 |     batch_outputs: List[Dict[str, Any]],
 |     batch_metrics: Dict[str, Any],
 |     epoch: int,
 |     batch_number: int,
 |     is_training: bool,
 |     is_master: bool
 | ) -> None

This callback hook is called after the end of each batch. This is equivalent to BatchCallback.

on_epoch#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def on_epoch(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     metrics: Dict[str, Any],
 |     epoch: int,
 |     is_master: bool
 | ) -> None

This callback hook is called after the end of each epoch. This is equivalent to EpochCallback.

on_end#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def on_end(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     metrics: Dict[str, Any],
 |     epoch: int,
 |     is_master: bool
 | ) -> None

This callback hook is called after the final training epoch. The epoch is passed as an argument.

batch#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def batch(self)

Construct a BatchCallback wrapper for this TrainCallback.

The cls.Batch type is created by the metaclass.

epoch#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def epoch(self)

Construct an EpochCallback wrapper for this instance.

The cls.Epoch type is created by the metaclass.

end#

class TrainerCallback(Registrable, metaclass=_TrainerCallbackMeta):
 | ...
 | def end(self)

Construct an EpochCallback wrapping the on_end end-of-training hook.

The cls.End type is created by the metaclass.

GradientDescentTrainer#

@Trainer.register("gradient_descent", constructor="from_partial_objects")
class GradientDescentTrainer(Trainer):
 | def __init__(
 |     self,
 |     model: Model,
 |     optimizer: torch.optim.Optimizer,
 |     data_loader: DataLoader,
 |     patience: Optional[int] = None,
 |     validation_metric: str = "-loss",
 |     validation_data_loader: DataLoader = None,
 |     num_epochs: int = 20,
 |     serialization_dir: Optional[str] = None,
 |     checkpointer: Checkpointer = None,
 |     cuda_device: Optional[Union[int, torch.device]] = None,
 |     grad_norm: Optional[float] = None,
 |     grad_clipping: Optional[float] = None,
 |     learning_rate_scheduler: Optional[LearningRateScheduler] = None,
 |     momentum_scheduler: Optional[MomentumScheduler] = None,
 |     tensorboard_writer: TensorboardWriter = None,
 |     moving_average: Optional[MovingAverage] = None,
 |     batch_callbacks: List[BatchCallback] = None,
 |     epoch_callbacks: List[EpochCallback] = None,
 |     end_callbacks: List[EpochCallback] = None,
 |     trainer_callbacks: List[TrainerCallback] = None,
 |     distributed: bool = False,
 |     local_rank: int = 0,
 |     world_size: int = 1,
 |     num_gradient_accumulation_steps: int = 1,
 |     use_amp: bool = False
 | ) -> None

A trainer for doing supervised learning with gradient descent. It just takes a labeled dataset and a DataLoader, and uses the supplied Optimizer to learn the weights for your model over some fixed number of epochs. You can also pass in a validation dataloader and enable early stopping. There are many other bells and whistles as well.

Registered as a Trainer with the name "gradient_descent" (and is also the default Trainer). The constructor that is registered is from_partial_objects - see the arguments to that function for the exact keys that should be used, if you are using a configuration file. They largely match the arguments to __init__, and we don't repeat their docstrings in from_partial_objects.

Parameters

model : Model
An AllenNLP model to be optimized. Pytorch Modules can also be optimized if their forward method returns a dictionary with a "loss" key, containing a scalar tensor representing the loss function to be optimized.

If you are training your model using GPUs, your model should already be on the correct device. (If you are using our train command this will be handled for you.)

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
optimizer : torch.nn.Optimizer
An instance of a Pytorch Optimizer, instantiated with the parameters of the model to be optimized.
data_loader : DataLoader
A DataLoader containing your Dataset, yielding padded indexed batches.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
patience : Optional[int] > 0, optional (default = None)
Number of epochs to be patient before early stopping: the training is stopped after patience epochs with no improvement. If given, it must be > 0. If None, early stopping is disabled.
validation_metric : str, optional (default = "-loss")
Validation metric to measure for whether to stop training using patience and whether to serialize an is_best model each epoch. The metric name must be prepended with either "+" or "-", which specifies whether the metric is an increasing or decreasing function.
validation_data_loader : DataLoader, optional (default = None)
A DataLoader to use for the validation set. If None, then use the training DataLoader with the validation data.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
num_epochs : int, optional (default = 20)
Number of training epochs.
serialization_dir : str, optional (default = None)
Path to directory for saving and loading model files. Models will not be saved if this parameter is not passed.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
checkpointer : Checkpointer, optional (default = None)
A Checkpointer is responsible for periodically saving model weights. If none is given here, we will construct one with default parameters.
cuda_device : int, optional (default = -1)
An integer specifying the CUDA device(s) to use for this process. If -1, the CPU is used. Data parallelism is controlled at the allennlp train level, so each trainer will have a single GPU.
grad_norm : float, optional (default = None)
If provided, gradient norms will be rescaled to have a maximum of this value.
grad_clipping : float, optional (default = None)
If provided, gradients will be clipped during the backward pass to have an (absolute) maximum of this value. If you are getting NaNs in your gradients during training that are not solved by using grad_norm, you may need this.
learning_rate_scheduler : LearningRateScheduler, optional (default = None)
If specified, the learning rate will be decayed with respect to this schedule at the end of each epoch (or batch, if the scheduler implements the step_batch method). If you use torch.optim.lr_scheduler.ReduceLROnPlateau, this will use the validation_metric provided to determine if learning has plateaued. To support updating the learning rate on every batch, this can optionally implement step_batch(batch_num_total) which updates the learning rate given the batch number.
momentum_scheduler : MomentumScheduler, optional (default = None)
If specified, the momentum will be updated at the end of each batch or epoch according to the schedule.
tensorboard_writer : TensorboardWriter, optional
If this is not provided, we will construct a TensorboardWriter with default parameters and use that.
moving_average : MovingAverage, optional (default = None)
If provided, we will maintain moving averages for all parameters. During training, we employ a shadow variable for each parameter, which maintains the moving average. During evaluation, we backup the original parameters and assign the moving averages to corresponding parameters. Be careful that when saving the checkpoint, we will save the moving averages of parameters. This is necessary because we want the saved model to perform as well as the validated model if we load it later. But this may cause problems if you restart the training from checkpoint.
batch_callbacks : List[BatchCallback], optional (default = None)
A list of callbacks that will be called at the end of every batch, during both train and validation.
epoch_callbacks : List[EpochCallback], optional (default = None)
A list of callbacks that will be called at the end of every epoch, and at the start of training (with epoch = -1).
end_callbacks : List[EpochCallback], optional (default = None)
A list of callbacks that will be called after the final epoch at the end of training. The type of the callbacks is the same as epoch_callbacks.
trainer_callbacks : List[TrainerCallback], optional (default = None)
A list of callbacks that will be called at each batch, epoch, and at the start and end of training.
distributed : bool, optional (default = False)
If set, PyTorch's DistributedDataParallel is used to train the model in multiple GPUs. This also requires world_size to be greater than 1.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately (you need a top-level "distributed" key, next to the "trainer" entry, that specifies a list of "cuda_devices").
local_rank : int, optional (default = 0)
This is the unique identifier of the Trainer in a distributed process group. The GPU device id is used as the rank.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
world_size : int, optional (default = 1)
The number of Trainer workers participating in the distributed training.

In a typical AllenNLP configuration file, this parameter does not get an entry under the "trainer", it gets constructed separately.
num_gradient_accumulation_steps : int, optional (default = 1)
Gradients are accumulated for the given number of steps before doing an optimizer step. This can be useful to accommodate batches that are larger than the RAM size. Refer Thomas Wolf's post for details on Gradient Accumulation.
use_amp : bool, optional (default = False)
If True, we'll train using Automatic Mixed Precision.

rescale_gradients#

class GradientDescentTrainer(Trainer):
 | ...
 | def rescale_gradients(self) -> float

Performs gradient rescaling. Is a no-op if gradient rescaling is not enabled.

Returns the norm of the gradients.

batch_outputs#

class GradientDescentTrainer(Trainer):
 | ...
 | def batch_outputs(
 |     self,
 |     batch: TensorDict,
 |     for_training: bool
 | ) -> Dict[str, torch.Tensor]

Does a forward pass on the given batch and returns the output dictionary that the model returns, after adding any specified regularization penalty to the loss (if training).

train#

class GradientDescentTrainer(Trainer):
 | ...
 | def train(self) -> Dict[str, Any]

Trains the supplied model with the supplied parameters.

get_checkpoint_state#

class GradientDescentTrainer(Trainer):
 | ...
 | @contextmanager
 | def get_checkpoint_state(
 |     self
 | ) -> Iterator[Tuple[Dict[str, Any], Dict[str, Any]]]

from_partial_objects#

class GradientDescentTrainer(Trainer):
 | ...
 | @classmethod
 | def from_partial_objects(
 |     cls,
 |     model: Model,
 |     serialization_dir: str,
 |     data_loader: DataLoader,
 |     validation_data_loader: DataLoader = None,
 |     local_rank: int = 0,
 |     patience: int = None,
 |     validation_metric: str = "-loss",
 |     num_epochs: int = 20,
 |     cuda_device: Optional[Union[int, torch.device]] = None,
 |     grad_norm: float = None,
 |     grad_clipping: float = None,
 |     distributed: bool = False,
 |     world_size: int = 1,
 |     num_gradient_accumulation_steps: int = 1,
 |     use_amp: bool = False,
 |     no_grad: List[str] = None,
 |     optimizer: Lazy[Optimizer] = Lazy(Optimizer.default),
 |     learning_rate_scheduler: Lazy[LearningRateScheduler] = None,
 |     momentum_scheduler: Lazy[MomentumScheduler] = None,
 |     tensorboard_writer: Lazy[TensorboardWriter] = Lazy(TensorboardWriter),
 |     moving_average: Lazy[MovingAverage] = None,
 |     checkpointer: Lazy[Checkpointer] = Lazy(Checkpointer),
 |     batch_callbacks: List[BatchCallback] = None,
 |     epoch_callbacks: List[EpochCallback] = None,
 |     end_callbacks: List[EpochCallback] = None,
 |     trainer_callbacks: List[TrainerCallback] = None
 | ) -> "Trainer"

This method exists so that we can have a documented method to construct this class using FromParams. If you are not using FromParams or config files, you can safely ignore this method.

The reason we can't just use __init__ with FromParams here is because there are sequential dependencies to this class's arguments. Anything that has a Lazy[] type annotation needs something from one of the non-Lazy arguments. The Optimizer needs to have the parameters from the Model before it's constructed, and the Schedulers need to have the Optimizer. Because of this, the typical way we construct things FromParams doesn't work, so we use Lazy to allow for constructing the objects sequentially.

If you're not using FromParams, you can just construct these arguments in the right order yourself in your code and call the constructor directly.

trainer

Trainer#

default_implementation#

train#

get_checkpoint_state#

BatchCallback#

__call__#

TensoboardBatchMemoryUsage#

__call__#

EpochCallback#

__call__#

TrackEpochCallback#

__call__#

TrainerCallback#

on_batch#

on_epoch#

on_end#

batch#

epoch#

end#

GradientDescentTrainer#

rescale_gradients#

batch_outputs#

train#

get_checkpoint_state#

from_partial_objects#

call#

call#

call#

call#