Rebased onto RWKV/RWKV7-G1k-7.2B-20260930 with some recent adapters added.
She's still quite base model ish / limited post-training, that hasn't changed. Still learning. :) Still RNN. Still pretty decent at raw language modeling.
I did have to patch a specific 96 token block of the lm_head that changed in the base model with the prior model's lm_head, which may mix up merges.