upvote
As was the case with GLM 5.3, it seems that there is still much juice to be squeezed from post-training
reply