v0.7.0

What's Changed

• Do not cast integers in in Axon.MixedPrecision.cast/2 by @jonatanklosko in #562
• Add support for global layer options by @jonatanklosko in #563
• Fix warnings by @jonatanklosko in #560
• improved doc formatting by fixing indentation and escape characters by @RicardoSantos-99 in #565
• Fix typos in docs about options by @preciz in #566
• Fix docs file basename by @preciz in #569
• Use model state struct instead of parameters by @seanmor5 in #553
• Add meta property to nodes by @seanmor5 in #571
• Fix iteration counts by @seanmor5 in #572
• Support blocks with multiple inputs by @seanmor5 in #574
• Fix missing state in training by @seanmor5 in #579
• Add inspect_values option by @seanmor5 in #581
• Use templates as parameters by @seanmor5 in #588
• Add rewrite_nodes function by @seanmor5 in #589
• Refactor containers by @seanmor5 in #590
• Provide layer name as hook name by @seanmor5 in #536
• Add simple quantization API by @seanmor5 in #586
• Improve language and fix function arity in docs by @preciz in #591
• Add time-series regression example notebook by @santiago-imelio in #587
• Update changelog with changes between v0.5.1..v0.6.0 by @preciz in #593
• Release v0.7.0 by @seanmor5 in #596

New Contributors

• @RicardoSantos-99 made their first contribution in #565
• @santiago-imelio made their first contribution in #587

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

v0.6.1

What's Changed

• test: ensure Loop.checkpoint reacts to event filter by @polvalente in #520
• Fix log-sum-exp to match the definition by @jonatanklosko in #527
• Fix typos and add codespell GitHub action by @kianmeng in #528
• Iteration bug multiple plots by @robinmonjo in #525
• refactor(examples): consistent CIFAR-10 shapes by @grzuy in #534
• ci: updates CI versions by @grzuy in #515
• Add block utility by @seanmor5 in #524
• Add metadata to layers by @seanmor5 in #535
• Fix missing reference in documentation. by @matiasgali in #539
• test: fixes Axon.block tests by @grzuy in #537
• Do not prefil metrics map by @josevalim in #440
• Fix omission in XOR example. by @matiasgali in #541
• Do not cast integer to float unintentionally by @seanmor5 in #547
• Update "Horses or humans" example. by @matiasgali in #543
• refactor(examples): attempt to improve clarity of where some hard-coded numbers coming from or mean by @grzuy in #512
• Introduce composite params, fix stability issue with unrolled RNNs by @seanmor5 in #550
• Force garbage collection option by @seanmor5 in #551
• Fix resize methods with kernel by @jonatanklosko in #554
• Add option to control anti-aliasing in the resize layer by @jonatanklosko in #555
• Fix event count for multiple handlers by @hwuethrich in #556
• Patch release for MLIR by @seanmor5 in #559

New Contributors

• @matiasgali made their first contribution in #539
• @hwuethrich made their first contribution in #556

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

v0.6.0

Prep for release 0.6 (#521)

v0.5.1

What's Changed

• Fix group normalization by @seanmor5 in #477

Full Changelog: v0.5.0...v0.5.1

Release v0.5.0

What's Changed

• Add more integration tests by @seanmor5 in #466
• Remove ONNX from CI by @seanmor5 in #462
• docs: update Axon.Loop docs by @polvalente in #469
• Remove deprecated transform by @seanmor5 in #467
• Prepare for Nx 0.5 by @seanmor5 in #471
• Deprecate Axon.Loop.handle/4 by @seanmor5 in #470
• docs: update channels default value mentioned in the docs by @nickgnd in #473
• Release v0.5 by @seanmor5 in #476

New Contributors

• @nickgnd made their first contribution in #473

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

v0.4.1

What's Changed

• Fix mismatch for certain optimizers with dropout by @seanmor5 in #458

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

v0.4.0

What's Changed

• Add gradient accumulation by @seanmor5 in #397
• Update notebook dependencies by @tcoyze in #398
• Return init params on compile by @josevalim in #400
• Update your_first_axon_model.livemd by @joelpaulkoch in #401
• Remove redundant cells by @msluszniak in #402
• Support lazy container inputs by @seanmor5 in #403
• Start an integration test suite by @seanmor5 in #404
• Unify event handlers in loops by @seanmor5 in #405
• Doc latex fixes by @chadykamar in #406
• Fix issue with initialization missing keys for some layers by @seanmor5 in #407
• Fix issue with namespaced stateful models by @seanmor5 in #409
• Training improvements by @seanmor5 in #411
• Add plot handler by @cigrainger in #412
• Training upgrades by @seanmor5 in #420
• Update schedules to require initial value by @seanmor5 in #421
• Make training deterministic by @seanmor5 in #424
• Use global event counts for correct filters in loops by @seanmor5 in #425
• Remove cast to Nx.Defn.Expr and simplify logits in losses by @josevalim in #429
• Build evaluator and zero metrics only once by @josevalim in #428
• Use a layer state to manage dropout state so it changes between runs by @seanmor5 in #426
• Provide deterministic builds by @josevalim in #427
• Use a more assertive name on plotting by @josevalim in #418
• Fix conv transpose with channels last by @seanmor5 in #431
• Fix Basic Examples by @nskins in #437
• Implement dropout as a passthrough by @josevalim in #436
• Only compute random key on init by @josevalim in #435
• Remove extra usage of 'expected input' from error message in get_input/3 by @nskins in #438
• Fix text generator example by @nskins in #441
• Add paragraph on integrating with Nx.Serving by @jonatanklosko in #442
• Include the folder for the guide links by @matthusby in #445
• Fix typo in Axon.Display.as_table by @NduatiK in #447
• Fix documentation for evaluator by @seanmor5 in #449
• Fix Fashion MNIST autoencoder example by @nskins in #439
• Fix issues with mismatched templates by @seanmor5 in #450
• Fix deprecation warnings with new Nx by @seanmor5 in #454
• Updates for release 0.4 by @seanmor5 in #456

New Contributors

• @tcoyze made their first contribution in #398
• @joelpaulkoch made their first contribution in #401
• @chadykamar made their first contribution in #406
• @matthusby made their first contribution in #445

Full Changelog: v0.3.0...v0.4.0