Comment by topwalktown

Transformers like Llama use rotary embeddings which are applied in every single attention layer

https://github.com/huggingface/transformers/blob/222505c7e4d...