FA2 and rotary positional embeddings

[benjamin-kroeger]

Hey thanks for building this framework it is exactly what I need for my project but I was wondering, whether there is a particular reason for why flash attention 2 and rotary positional embeddings were discarded from the standard Llama implementation?

[SeanLee97]

hi @benjamin-kroeger, it might be because this code is based on an earlier transformers version. I will upgrade it to be compatible with the latest transformers when I am free. PRs are also welcome!