I wish I could link it, but I had also read another strategy for reasoning effort, perhaps the most naive implementation, described as whether or not you ignored or accepted the end of sequence token, n-times, n being some count of subsequent ignores.
Now how true that is, I don't know. You could reasonably implement this locally with an inference provider, but I don't know any that do, or I haven't used them at least.
I may also have the technical detail incorrect, but it did come down to essentially "continuing inference beyond a natural termination."
Now how true that is, I don't know. You could reasonably implement this locally with an inference provider, but I don't know any that do, or I haven't used them at least.
I may also have the technical detail incorrect, but it did come down to essentially "continuing inference beyond a natural termination."