v0.8.0: tinygrad, small fixes and updates

TLDR

• tinygrad backend added
• resolve warning in py3.11 related to docstring
• remove graph break for unpack
• breaking TF layers were updated to follow new instructions, new layers compatible with TF 2.16, and not compatible with old TF (certainly does not work with TF2.13)

What's Changed

• Fix invalid escape sequence in einsum docstring by @atwam in #298
• Tinygrad support by @blueridanus in #297
• Coerce bool to int in unpack by @drubinstein in #287
• Remove oneflow from testing by @arogozhnikov in #289
• tests: fix torch installation to force CPU by @arogozhnikov in #288
• Allow anonymous axes in parse_shape, fix #302 by @arogozhnikov in #303
• Codebase standards + update TF layers by @arogozhnikov in #318
• update github actions by @arogozhnikov in #319

New Contributors

• @drubinstein made their first contribution in #287
• @atwam made their first contribution in #298
• @blueridanus made their first contribution in #297

Full Changelog: v0.7.0...v0.8.0

V0.7.0: torch.compile, preserve axis identity, array api

Major changes:

• torch.compile just works, registration of operations happens automatically
• JAX's distributed arrays can use ellipses, and in general ellipsis processing now preserves axis identity. This involved changing internal gears of einops.
• Array API: einops operations can be used with any framework that follows the standard (see einops.array_api)
• Python 3.7 is dead. Good bye, you were great at the time
• Gluon is dropped as previously announced
• reduce/repeat/rearrange all accept lists now

PRs list

• Preserve axis identity + drop python 3.7 by @arogozhnikov in #255
• Support array api by @arogozhnikov in #261
• add type-based caching by @arogozhnikov in #262
• Uniform support for list inputs across rearrange/reduce/repeat by @arogozhnikov in #263
• Drop gluon by @arogozhnikov in #264
• automatically register torch ops in torchdynamo by @arogozhnikov in #265
• release 0.7.0rc1, highlight changes in README by @arogozhnikov in #266
• cover dynamic shapes in torch.compile, introduce fallback if shape was not cacheable by @arogozhnikov in #275
• bump version by @arogozhnikov in #276
• fix #279, update description by @arogozhnikov in #281
• Add support for any/all reductions in einops.reduce by @arogozhnikov in #282
• maintenance: cache pip dependencies by @arogozhnikov in #283
• maintenance: update versions of github actions by @arogozhnikov in #285

Full Changelog: v0.6.1...v0.7.0

V0.7.0rc2: allow dynamic shapes in torch.compile

What's Changed

• cover dynamic shapes in torch.compile, introduce fallback if shape was not cacheable by @arogozhnikov in #275
• bump version by @arogozhnikov in #276

Full Changelog: v0.7.0rc1...v0.7.0rc2

V0.7.0rc1: torch.compile, preserve axis identity, array api

Major changes:

• torch.compile just works, registration of operations happens automatically
• JAX's distributed arrays can use ellipses, and in general ellipsis processing now preserves axis identity. This involved changing internal gears of einops.
• Array API: einops operations can be used with any framework that follows the standard (see einops.array_api)
• Python 3.7 is dead. Good bye, you were great at the time
• Gluon is dropped as previously announced
• Reduce/repeat/rearrange all accept lists now

What's Changed

• Support array api by @arogozhnikov in #261
• add type-based caching by @arogozhnikov in #262
• Uniform support for list inputs across rearrange/reduce/repeat by @arogozhnikov in #263
• Drop gluon by @arogozhnikov in #264
• automatically register torch ops in torchdynamo by @arogozhnikov in #265
• release 0.7.0rc1, highlight changes in README by @arogozhnikov in #266
• Preserve axis identity + drop python 3.7 by @arogozhnikov in #255

Full Changelog: v0.6.1...v0.7.0rc1

V0.6.2rc0: drop python 3.7 + preserve axis identity

pre-release is published to allow public testing of new caching logic (pattern analysis is now dependent on input dimensionality to preserve axis identity).

What's Changed

• Preserve axis identity + drop python 3.7 by @arogozhnikov in #255

Full Changelog: v0.6.1...v0.6.2rc0

V0.6.1: support paddle, support pytorch.compile

• einops layers perfectly interplay with torch.compile
• einops operations needs registration: run einops._torch_specific.allow_ops_in_compiled_graph() before torch.compile
• paddle is now supported (thanks to @zhouwei25)
• as previously announced, support of mxnet is dropped

What's Changed

• Add PaddlePaddle backend for einops by @zhouwei25 in #242 , #245
• Add allow_ops_in_compiled_graph to support torch.compile by @arogozhnikov in #251
• torch.concat -> torch.cat to support packing for torch < 1.10 by @arogozhnikov in #238
• chores: improve documentation and testing by @arogozhnikov in #246
• Remove mxnet by @arogozhnikov in #231

New Contributors

• @zhouwei25 made their first contribution in #242

Full Changelog: v0.6.0...v0.6.1

V0.6.0: pack and unpack

What's Changed

• Introduce einops.pack and einops.unpack by @arogozhnikov in #222
• Update example to match description by @EPronovost in #217
• Improve type hinting by @arogozhnikov in #221
• Cosmetics for pack/unpack: documentation and comments by @arogozhnikov in #223
• Preparations for 0.6.0 release by @arogozhnikov in #224

New Contributors

• @EPronovost made their first contribution in #217

Announcement

Sunsetting experimental mxnet support: no demand and package is outdated, with numerous deprecations and poor support of corner cases. 0.6.0 will be the last release with mxnet backend.

Full Changelog: v0.5.0...v0.6.0

Einops v0.5.0: einsum + support of flax and oneflow.

What's Changed

• Create einsum operation by @MilesCranmer in #197
• Add flax layers by @arogozhnikov in #214
• Add oneflow backend by @arogozhnikov in #181
• Add oneflow backend by @rentainhe in #180
• Fix wrong error message by @arogozhnikov in #196
• Clarify documentation re. default bias on EinMix by @maxeonyx in #201
• corrected spelling mistake: einsops -> einops by @cs-mshah in #205
• add mean-reduction for bfloat16, fix #206 by @arogozhnikov in #209
• add py.typed (adopt PEP 561) by @arogozhnikov in #211
• Delete tensorflow-specific readme file by @arogozhnikov in #212
• Adopt pypa/hatch by @arogozhnikov in #213

New Contributors

• @rentainhe made their first contribution in #180
• @MilesCranmer made their first contribution in #197
• @maxeonyx made their first contribution in #201
• @cs-mshah made their first contribution in #205

Full Changelog: v0.4.1...v0.5.0

Einops v0.4.1: Avoid importing numpy if it is not required

What's Changed

• fix numpy dependency problem by @lucidrains in #176

New Contributors

• @lucidrains made their first contribution in #176

Full Changelog: v0.4.0...v0.4.1

Einops v0.4.0: EinMix and torch.jit.script

Main Changes

• torch.jit.script is supported (in addition to previous torch.jit.trace)
• EinMix (swiss-knife for next-gen MLPs) is added. A much-improved einsum/linear layer is now available.
• einops.repeat in torch does not create copy when possible

Detailed PRs

• Update documentation by @arogozhnikov in #137
• Multiple updates in docs, add Rearrange layer to torch test by @arogozhnikov in #138
• Add support for torch scripting of einops layers by @arogozhnikov in #139
• Introduce EinMix - swiss-knife for next-gen MLPs by @arogozhnikov in #142
• Docs improvements: wording, visual style, EinMix by @arogozhnikov in #143
• Move docs to a separate folder by @arogozhnikov in #144
• Type hinting + add testing for EinMix composition/decomposition by @arogozhnikov in #154
• Reject repeated axes in parse_shape by @dmitriy-serdyuk in #159
• Enable ellipsis in patterns for parse_shape. by @dmitriy-serdyuk in #162

New Contributors

• @dmitriy-serdyuk made their first contribution in #159

Full Changelog: v0.3.2...v0.4.0