Interesting thought, I did a very quick search over the transcripts and it seems models mention this as a reason for not-paging in a good chunk of transcripts, mostly Luna and Deepseek.
Seems like it would be an interesting ablation to run. I’ve made the repo public if you’re interested in forking it and playing around.
Interesting thought, I did a very quick search over the transcripts and it seems models mention this as a reason for not-paging in a good chunk of transcripts, mostly Luna and Deepseek.
Seems like it would be an interesting ablation to run. I’ve made the repo public if you’re interested in forking it and playing around.