Model taking a copy of its weights, rewriting them, and in-place replacement of the weights file?
replyThis sounds like RL. Iteration with diminishing marginal returns.
replyIf you automated the post-training I can see that you would progress faster, but just reach the same limit sooner.