1 comment

[ 6.2 ms ] story [ 88.3 ms ] thread
A new post-pretraining method for LLMs with an expansion of Transformer blocks.