Fantastic!
Great model! It generating a lot of details and descriptions, also following context! It gave a new life to all my character cards! I'm using oobabooga with Liminal Drift preset.
Thank you!
I have just released a v1.2 version,, it takes a little more care to make it keep the same verbosity, but it's much less likely to go off the walls and start ignoring instructions and go schizophrenic
Unfortunately, I can only use the 4 bit version, if The Bloke releases a 4 bit version of the new model, I'll be happy to try it out.
I'm quantizing the gptQ one right now, it's just going slow as usual
I tested the 4 bit version of the 1.2 model for a little, and for now i feel it like she got worse with card's which used w++ syntax - it has become less descriptive and confusing in events and roles, inventing non-existent past events or assigning itself a role prescribed for the user. But in terms of the execution of the instruction, it has become much better!
But it is too early to draw final conclusions, it is necessary to test more thoroughly. I'm using 1.0 version from "TheBloke/Huginn-13B-GPTQ".
yeah, it does sadly take a bit more effort to make it be as verbose..