// the find
bojone/attention
some attention implements
A 2019 implementation of the attention mechanism from 'Attention is All You Need', targeting Python 2.7 + TF 1.8 + Keras 2.2.4. Two files: one for Keras, one for raw TensorFlow. The author explicitly abandoned it in favor of bert4keras.
The code is minimal and readable — good for understanding the raw mechanics without framework noise. The author (Su Jianlin / bojone) is a respected ML researcher in the Chinese community with a well-known blog post explaining the implementation. The bert4keras pointer in the README is actually useful — it tells you exactly where to find the maintained version.
Abandoned in 2019 and pinned to Python 2.7 and TF 1.8, both of which are long dead. Two files and a README is not a library. No tests, no packaging, no documentation beyond a Chinese blog link. If you need attention in 2024, PyTorch has it built in, transformers has it, and even the author's own bert4keras has superseded this. There is no reason to adopt this repo.