AttentionSmithy v1.3.0

What's Changed

• Update dev branch with new custom embedding classes by @CCranney in #39
• Update dev branch from main by @CCranney in #41
• linformer initialized by @CCranney in #42
• Dev/expand norm and activations by @CCranney in #43
• longformer, linformer, normalization modularity, extended activation functions by @CCranney in #44

Full Changelog: v1.2.1...v1.3.0

AttentionSmithy v1.2.1

What's Changed

• rotary custom embedding dimension error fix by @CCranney in #40

Full Changelog: v1.2.0...v1.2.1

AttentionSmithy v1.2.0

What's Changed

• adding custom rotary embeddings and continuous value embeddings by @CCranney in #38

Full Changelog: v1.1.1...v1.2.0

AttentionSmithy v1.1.1

What's Changed

• update encoder/decoder freeze layers to run .eval() on frozen layers by @CCranney in #37

Full Changelog: v1.1.0...v1.1.1

AttentionSmithy v1.1.0

Began implementing specialized transformer components (in this case relating to the Perceiver architecture).

What's Changed

• Machine translation example fix by @CCranney in #33
• updating example script README.md with environment information by @CCranney in #34
• implementing perceiver classes by @CCranney in #36

Full Changelog: v1.0.3...v1.1.0

AttentionSmithy v1.0.3

Generator strategy bug fix, cleaning up unmaintained Big Bird implementation.

Note:

Future Big Bird development will require a complete revamp, and thus this old code is unusable. The old code is archived under the archive/big-bird branch.

What's Changed

• fix - generator strategy no_repeat_ngram_size=1 now blocks repeat tokens by @CCranney in #29
• workflow check update - increment only required when src is updated by @CCranney in #31
• big bird code archived, removing from main by @CCranney in #32

Full Changelog: v1.0.1...v1.0.3

v1.0.1

Has fixes to the project name metadata to make it pip installable.

What's Changed

• update project name for pypi readability and installation by @CCranney in #27

Full Changelog: v1.0.0...v1.0.1

v1.0.0

First official release of AttentionSmithy!

What's Changed

• generator classes implemented by @CCranney in #2
• Multi gpu adjustments made by @CCranney in #17
• Cleanup generator classes by @CCranney in #18
• catches error for unrecognized generation strategy by @CCranney in #19
• alibi custom embedding can be ignored for specific values (to avoid a… by @CCranney in #23
• add get_available_gpu_count function to AttentionSmithy utils.py by @CCranney in #24
• refactored numeric embedding manager class (and all related subclasse… by @CCranney in #25
• Documentation prep by @CCranney in #26

Full Changelog: https://github.com/xomicsdatascience/AttentionSmithy/commits/v1.0.0