f-dangel / backpack

98%

master: 98%

LAST BUILD ON BRANCH multiple-inputs

branch: multiple-inputs

CHANGE BRANCH

x

Reset

• multiple-inputs
• 1.1.0
• 1.1.1
• 1.2.0
• adaptive-pooling-for-diagh
• automate-pypi-release
• batch-grad-transform
• bce-with-logits-loss
• bce_kfac
• black-22.1.0
• branching
• broken-kfra-conv
• bug-convolution-groups
• changelog-update-1-6
• ci-test-multiple-pytorch-versions
• conv-batch-grad-extensions
• conv-tests
• conv-transpose
• conv-weight-mjp-functorch
• conv1d-3d-derivatives
• convnd-extract-weight-diagonal-support-groups
• convnd-weight-jac-support-groups
• convndtranspose-extract-weight-diagonal-support-groups
• convndtranspose-weight-jac-support-groups
• crossentropy-unsupported-parameters
• curvmatprod-example-cg-newton
• curvmatprod-example-hutchinson
• deprecate-python35
• deprecate-system-packages
• development
• development-backup-rnn-merge
• diag-ggn-parent-class
• doc-add-per-sample-curvatures
• doc-update-supported-layers
• docstring-standards
• eingroup_improvement
• extension-hook
• first-order-resnets
• fix-batch-size-1-settings
• fix-example-imports
• fix-examples-website
• fix-flake8-abc
• fix-gpu-tests
• fix-pypi-release
• fix-pytorch-memlab
• fix-rtd-build
• fix-second-order-extensions-elu-selu
• forbid-inplace
• generalize-jac-t-interface
• hbp-no-support-groups
• hvp-allow-gradients
• improve-backpropagation-conditions
• improve-batch-axis-detection
• improve-doc-coverage
• improve-readme
• inconsistent-yield
• issue30
• kfac-ConvNd
• linear-support-additional-dimensions
• linear-support-additional-dimensions-extensions
• lint
• lint-doc-examples
• lint-examples
• lstm-proj-size
• martens2010estimating
• master
• mc-samples
• more-examples
• mse-loss-hyperparameters
• mse-with-vectors
• new-shape-convention
• no-io-context
• owkin-benchmark
• padding-custom-module
• partial-docstring-checks
• pep8-style
• pfischer-hbp-batchnorm
• post-extension-hook
• prepare-merge-development-into-rnn
• prepare-merge-rnn-into-development
• prepare-release
• pytorch-1-13-support
• refactor-and-update-ggnmp
• refactor-curvmatprod
• release
• remove-detach
• remove-get-module
• remove-pch-med
• remove_opt_einsum
• remove_shape_buffers
• resnet-cifar10
• restructure-cmp
• retain-graph-example
• revert-save-memory
• rnn
• rtd
• rtd-logo
• rtd-web
• save-memory-use-case
• second-order-losses
• simplify-core-tests
• slicing-custom-module
• sqrt-ggn-batchnorm
• sqrt-ggn-resnet-modules
• sqrtggnexact-extension
• sqrtggnmc-extension
• subsampling
• subsampling-batch-grad
• subsampling-bias-jac-t
• subsampling-example
• subsampling-jac-t-mat-prod
• subsampling-rnn-lstm-param-jac-t
• subsampling-sqrt-ggn
• subsampling-sqrt-hessian
• subsampling-weight-jac-t
• support-torch-2
• test-case-flatten-no-op
• test-coverage
• test-derivatives
• test-extensions
• test-hessian-properties
• test-reduce-run-time
• torch-1-12-0-compatibility
• torch-compatibility
• try-view-unnecessary-from-torch-1.4
• update-changelog
• update-ci
• update-doc
• update-type-annotations-python-37
• use-case-batched-jacobians
• use-cases
• use-isort
• use-unfoldNd-package
• writing-new-layers

[REF] Decompose tuple-valued quantities in backprop for multi-input modules