There are some already (one of the openllamas I think, and that discord group made one with the linear attention recently although trained on fewer params I thi

There are some already (one of the openllamas I think, and that discord group made one with the linear attention recently although trained on fewer params I think) Maybe not easy to switch to a non-meta-llama ?