model

superduper.ext.torch.model

Source code

create_batch

create_batch(args)

Parameter	Description
args	single data point for batching

Create a singleton batch in a manner similar to the PyTorch dataloader.

create_batch(3.).shape
# torch.Size([1])
x, y = create_batch([torch.randn(5), torch.randn(3, 7)])
x.shape
# torch.Size([1, 5])
y.shape
# torch.Size([1, 3, 7])
d = create_batch(({'a': torch.randn(4)}))
d['a'].shape
# torch.Size([1, 4])

torchmodel

torchmodel(class_obj)

Parameter	Description
class_obj	Class to decorate

A decorator to convert a torch.nn.Module into a TorchModel.

Decorate a torch.nn.Module so that when it is invoked, the result is a TorchModel.

unpack_batch

unpack_batch(args)

Parameter	Description
args	a batch of model outputs

Unpack a batch into lines of tensor output.

unpack_batch(torch.randn(1, 10))[0].shape
# torch.Size([10])
out = unpack_batch([torch.randn(2, 10), torch.randn(2, 3, 5)])
type(out)
# <class 'list'>
len(out)
# 2
out = unpack_batch({'a': torch.randn(2, 10), 'b': torch.randn(2, 3, 5)})
[type(x) for x in out]
# [<class 'dict'>, <class 'dict'>]
out[0]['a'].shape
# torch.Size([10])
out[0]['b'].shape
# torch.Size([3, 5])
out = unpack_batch({'a': {'b': torch.randn(2, 10)}})
out[0]['a']['b'].shape
# torch.Size([10])
out[1]['a']['b'].shape
# torch.Size([10])

TorchModel

TorchModel(self,
     identifier: str,
     db: dataclasses.InitVar[typing.Optional[ForwardRef('Datalayer')]] = None,
     uuid: str = None,
     *,
     preferred_devices: 't.Sequence[str]' = ('cuda',
     'mps',
     'cpu'),
     device: 't.Optional[str]' = None,
     trainer: 't.Optional[Trainer]' = None,
     artifacts: 'dc.InitVar[t.Optional[t.Dict]]' = None,
     signature: 'Signature' = '*args,
    **kwargs',
     datatype: 'EncoderArg' = None,
     output_schema: 't.Optional[Schema]' = None,
     flatten: 'bool' = False,
     model_update_kwargs: 't.Dict' = None,
     predict_kwargs: 't.Dict' = None,
     compute_kwargs: 't.Dict' = None,
     validation: 't.Optional[Validation]' = None,
     metric_values: 't.Dict' = None,
     object: 'torch.nn.Module',
     preprocess: 't.Optional[t.Callable]' = None,
     preprocess_signature: 'Signature' = 'singleton',
     postprocess: 't.Optional[t.Callable]' = None,
     postprocess_signature: 'Signature' = 'singleton',
     forward_method: 'str' = '__call__',
     forward_signature: 'Signature' = 'singleton',
     train_forward_method: 'str' = '__call__',
     train_forward_signature: 'Signature' = 'singleton',
     train_preprocess: 't.Optional[t.Callable]' = None,
     train_preprocess_signature: 'Signature' = 'singleton',
     collate_fn: 't.Optional[t.Callable]' = None,
     optimizer_state: 't.Optional[t.Any]' = None,
     loader_kwargs: 't.Dict' = None) -> None

Parameter	Description
identifier	Identifier of the leaf.
db	Datalayer instance.
uuid	UUID of the leaf.
artifacts	A dictionary of artifacts paths and DataType objects
signature	Model signature.
datatype	DataType instance.
output_schema	Output schema (mapping of encoders).
flatten	Flatten the model outputs.
model_update_kwargs	The kwargs to use for model update.
predict_kwargs	Additional arguments to use at prediction time.
compute_kwargs	Kwargs used for compute backend job submit. Example (Ray backend): compute_kwargs = dict(resources=...).
validation	The validation Dataset instances to use.
metric_values	The metrics to evaluate on.
object	Torch model, e.g. torch.nn.Module
preprocess	Preprocess function, the function to apply to the input
preprocess_signature	The signature of the preprocess function
postprocess	The postprocess function, the function to apply to the output
postprocess_signature	The signature of the postprocess function
forward_method	The forward method, the method to call on the model
forward_signature	The signature of the forward method
train_forward_method	Train forward method, the method to call on the model
train_forward_signature	The signature of the train forward method
train_preprocess	Train preprocess function, the function to apply to the input
train_preprocess_signature	The signature of the train preprocess function
collate_fn	The collate function for the dataloader
optimizer_state	The optimizer state
loader_kwargs	The kwargs for the dataloader
trainer	Trainer object to train the model
preferred_devices	The order of devices to use
device	The device to be used

Torch model. This class is a wrapper around a PyTorch model.

BasicDataset

BasicDataset(self,
     items,
     transform,
     signature)

Parameter	Description
items	items, typically documents
transform	function, typically a preprocess function
signature	signature of the transform function

Basic database iterating over a list of documents and applying a transformation.