A description for what each one does?
Hi, is there a discord to discuss these? I tried the simple one so far and liked the results, would like to know more about what to expect from each vector.
Hi, thanks for the feedback!
is there a discord to discuss these?
Probably the BeaverAI discord? You can find the link on any of TheDrummer's model cards.
I tried the simple one so far and liked the results
Great! The hybrid-reasoning system in GLM-4.6 made these tricky to train, so feedback is helpful.
would like to know more about what to expect from each vector
Yeah, I'm still figuring out model card best practices -_-! There are a few examples in these (more recent) model cards:
Also check out JukOfYork's jukofyork/creative-writing-control-vectors-v3.0.
That model card is dense, but it's the definitive technical explanation of how these work, why the llama.cpp built-in trainer isn't enough, etc.
The discussions there are probably the longest threads on HuggingFace!