Discovered pi-vcc, why pi-blackhole?

So literally two days ago I discovered pi-vcc from someone's comment reply in this sub. I installed it, and we're off to the races. Sub-second compaction with recall so it's near-lossless, great! happens without a model, even better! Then comes along pi-blackhole. It sounds like pi-vcc, but uses three separate LLMs (albeit small ones) to build better recall? Or something? Seems like a lot of work to move away from non-deterministic LLM generated summaries right back to calling models again. Anyone tried both? Worth the switch? I'm liking pi-vcc so far.

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论