A PyTorch Library for Meta-learning Research
Like 0.2.0 but with qpth made optional.
l2l.nn.MetaModule
and l2l.nn.ParameterTransform
for parameter-efficient finetuning.l2l.nn.freeze
and l2l.nn.unfreeze
.mkdocstrings
instead of pydoc-markdown
.text/news_topic_classification.py
example.detach_module
. (Nimish Sanghi)l2l.vision.models.get_pretrained_backbone()
.keep_requires_grad
flag to detach_module
. (Zhaofeng Wu)l2l.nn.Scale
.train_loss
logging in LightningModule
implementations with PyTorch-Lightning 1.5.RandomClassRotation
(https://github.com/learnables/learn2learn/pull/283) to incorporate multi-channelled inputs. (Varad Pimpalkhute)maml.py
and meta-sgd.py
and add tests to maml_test.py
and metasgd_test.py
to check for possible future memory leaks. (https://github.com/learnables/learn2learn/issues/284) (Kevin Zhang)l2l.data.EpisodicBatcher
.l2l.nn.PrototypicalClassifier
and l2l.nn.SVMClassifier
.l2l.vision.models.WRN28
.CNN4Backbone
, ResNet12Backbone
, WRN28Backbones
w/ pretrained weights.l2l.data.OnDeviceDataset
and implement device
parameter for benchmarks.l2l.data.partition_task
and l2l.data.InfiniteIterator
.ResNet12
.l2l.nn.KroneckerLinear
. (@timweiland)FilteredMetaDatasest
filter the classes used to sample tasks.UnionMetaDatasest
to get the union of multiple MetaDatasets.MiniImageNetCNN
to CNN4
and add embedding_size
argument.l2l.vision.models.ResNet12
l2l.vision.datasets.DescribableTextures
l2l.vision.datasets.Quickdraw
l2l.vision.datasets.FGVCFungi
labels_to_indices
and indices_to_labels
as optional arguments to l2l.data.MetaDataset
.l2l.vision.benchmarks
interface.l2l.optim
. (including l2l.optim.LearnableOptimizer
for meta-descent)l2l.algorithms.GBML
.nn.Modules
in l2l.nn
.l2l.update_module
as a more general alternative to l2l.algorithms.maml_update
.