Tags · lucidrains/local-attention

1.11.1

release more efficient look_around thanks to @minh-nguyenhoang !

hyper connect the local transformer

allow for inserting layers before the local attention, for testing a …

…new neural memory paper

remove a warning

make sure kv caching with dynamic pos bias works for local transformer

remove adding staticmethod

oops

revert back to using plain apply_rotary_pos_emb fn

allow for argmax sampling

PreviousNext