Abstraction framework on top of candle, viable embedding and attention layers

Hi candle team, I am working on a candle abstraction framework which is almost comparable with Keras. See here: https://github.com/BDUG/Lighter

I am currently working on the embedding and attention layers. While doing so, I dived through your code and derived the following, possible layers:

- Embedding: https://github.com/BDUG/Lighter/blob/main/docs/embedding.MD
- Attention: https://github.com/BDUG/Lighter/blob/main/docs/attention.MD

Anything I missed out?

Thanks in advance

Start by reviewing the linked Lighter embedding.MD and attention.MD documents, then compare their proposed layers with Candle's current capabilities. A useful outcome would require a concrete list of missing Candle functionality, defined behavior, and tests; the issue does not currently specify an implementation target.

Written by the indexing model from the issue text.