composer.core.types#

Reference for common types used throughout the composer library.

composer.core.types.Model#

Alias for torch.nn.Module.

Type: Module

composer.core.types.ModelParameters#

Type alias for model parameters used to initialize optimizers.

Type: Iterable[Tensor] | Iterable[Dict[str, Tensor]]

composer.core.types.Tensor#

Alias for torch.Tensor.

Type: Tensor

composer.core.types.Tensors#

Commonly used to represent e.g. a set of inputs, where it is unclear whether each input has its own tensor, or if all the inputs are concatenated in a single tensor.

Type: Tensor | Tuple[Tensor, …] | List[Tensor]

composer.core.types.Batch#

Union type covering the most common representations of batches. A batch of data can be represented in several formats, depending on the application.

Type: BatchPair | BatchDict | Tensor

composer.core.types.BatchPair#

Commonly used in computer vision tasks. The object is assumed to contain exactly two elements, where the first represents inputs and the second represents targets.

Type: Tuple[Tensors, Tensors] | List[Tensor]

composer.core.types.BatchDict#

Commonly used in natural language processing tasks.

Type: Dict[str, Tensor]

composer.core.types.Metrics#

Union type covering common formats for representing metrics.

Type: Metric | MetricCollection

composer.core.types.Optimizer[source]#

Alias for torch.optim.Optimizer

Type: Optimizer

composer.core.types.Optimizers#

Union type for indeterminate amounts of optimizers.

Type: Optimizer | List[Optimizer] | Tuple[Optimizer, …]

composer.core.types.PyTorchScheduler#

Alias for base class of learning rate schedulers such as torch.optim.lr_scheduler.ConstantLR

Type: torch.optim.lr_scheduler._LRScheduler

composer.core.types.Scaler#

Alias for torch.cuda.amp.GradScaler.

Type: torch.cuda.amp.grad_scaler.GradScaler

composer.core.types.JSON#

JSON Data

Type: str | float | int | None | List[’JSON’] | Dict[str, ’JSON’]

composer.core.types.Evaluators#

Union type for indeterminate amounts of evaluators.

Type: Many[Evaluator]

composer.core.types.StateDict#

pickale-able dict via torch.save()

Type: Dict[str, Any]

composer.core.types.Dataset[source]#

Alias for torch.utils.data.Dataset

Type: Dataset[Batch]

Functions

`as_batch_dict`	Casts a `Batch` as a `BatchDict`.
`as_batch_pair`	Casts a `Batch` as a `BatchPair`.

Classes

`DataLoader`	Protocol for custom DataLoaders compatible with `torch.utils.data.DataLoader`.
`MemoryFormat`	Enum class to represent different memory formats.
`_LRScheduler`	composer.core.types.torch.optim.lr_scheduler._LRScheduler
`GradScaler`	composer.core.types.torch.cuda.amp.grad_scaler.GradScaler

Exceptions

BreakEpochException

Raising this exception will immediately end the current epoch.

Attributes

Batch
BatchDict
BatchPair
Evaluators
JSON
Metrics
ModelParameters
Optimizers
StateDict
Tensors

exception composer.core.types.BreakEpochException[source]#

Bases: Exception

Raising this exception will immediately end the current epoch.

If you’re wondering whether you should use this, the answer is no.

class composer.core.types.DataLoader(*args, **kwargs)[source]#

Bases: Protocol

Protocol for custom DataLoaders compatible with torch.utils.data.DataLoader.

dataset#

Dataset from which to load the data.

Type: Dataset

batch_size#

How many samples per batch to load for a single device (default: 1).

Type: int, optional

num_workers#

How many subprocesses to use for data loading. 0 means that the data will be loaded in the main process.

Type: int

pin_memory#

If True, the data loader will copy Tensors into CUDA pinned memory before returning them.

Type: bool

drop_last#

If len(dataset) is not evenly divisible by batch_size, whether the last batch is dropped (if True) or truncated (if False).

Type: bool

timeout#

The timeout for collecting a batch from workers.

Type: float

sampler#

The dataloader sampler.

Type: Sampler[int]

prefetch_factor#

Number of samples loaded in advance by each worker. 2 means there will be a total of 2 * num_workers samples prefetched across all workers.

Type: int

class composer.core.types.MemoryFormat(value)[source]#

Bases: composer.utils.string_enum.StringEnum

Enum class to represent different memory formats.

See torch.torch.memory_format for more details.

CONTIGUOUS_FORMAT#: Default PyTorch memory format represnting a tensor allocated with consecutive dimensions sequential in allocated memory.

CHANNELS_LAST#: This is also known as NHWC. Typically used for images with 2 spatial dimensions (i.e., Height and Width) where channels next to each other in indexing are next to each other in allocated memory. For example, if C[0] is at memory location M_0 then C[1] is at memory location M_1, etc.

CHANNELS_LAST_3D#: This can also be referred to as NTHWC. Same as CHANNELS_LAST but for videos with 3 spatial dimensions (i.e., Time, Height and Width).

PRESERVE_FORMAT#: A way to tell operations to make the output tensor to have the same memory format as the input tensor.

composer.core.types.as_batch_dict(batch)[source]#

Casts a Batch as a BatchDict.

Parameters: batch (Batch) – A batch.
Raises: TypeError – If the batch is not a BatchDict.
Returns: BatchDict – The batch, represented as a BatchDict.

composer.core.types.as_batch_pair(batch)[source]#

Casts a Batch as a BatchPair.

Parameters: batch (Batch) – A batch.
Returns: BatchPair – The batch, represented as a BatchPair.
Raises: TypeError – If the batch is not a BatchPair.