.FastAI.RegistriesModelVariant
abstract type defined in module
FastAI.Registries
abstract type ModelVariant
A
ModelVariant handles loading a model, optionally with pretrained weights and transforming it so that it can be used for specific learning tasks.
are subblocks (see
issubblock) of
blocks = (xblock, yblock).
compatibleblocks
(variant) returns a tuple
(xblock, yblock) of
BlockLike
that are compatible with the model. This means that a variant can be used for a task with input and output blocks
blocks, if
issubblock
(blocks, compatibleblocks(variant)).
loadvariant
(::ModelVariant, xblock, yblock, checkpoint; kwargs...) loads a model compatible with block instances
xblock and
yblock, with (optionally) weights from
checkpoint.
The following page links back here: