Model starts to ramble in RP
I am fairly certain its my Text Completion presets, but not quite sure what to change. Do you have some defaults that I can load and work off from?
If you are using Context Shift - disable this. It will cause this issue.
Other issues:
- temp / rep pen setting.
- Increase rep pen first , 1.05+ ... RP may need 1.1+
- Then adjust temp.
Also note, this version can be used for RP ; but "rp" specific versions such as Dark Planet, Grand Gutenberg Darkness/Madness may be better suited as these
are far more stable with the cost of some creativity.
I could not find the option Context Shift. I did follow the Rep pen/ temp settings recommended, as I know that has quite a huge impact with models, so I always check those first.
Are you using Dynamic Temp?
I did enjoy the Gutenberg Models, but I wanted to try the other ones too, just to see which one suits me best.
Maybe to add to this, this is the part that confuses me:
I assume that the Instruct template is the one where I am supposed to put the template. However the options mentioned on Gutenberg like "Collapse Consecutive Newlines" are under Context template. However when I copy paste the template in the Story string, I get an error message that the format is not what it expected.
I am uncertain - based on screen shot - which "ai program" you are using.
Context shift may be under advanced options (guessing here).
RE: Input of a template.
Template should be used EXACTLY as is, including carriage returns (ie \n) . If these are missing or removed, it will comprise model operation.
You may want to try lmstudio.ai ; at this might be easier to operate.
The screenshot is from SillyTavern, here is the full overview:
I am using LM Studio as my "server", and Silly Tavern to deal with Character cards.
It looks like you have everything correct here ; suggest trying Dark Planet 8B (or new Dark Planet 8b L3.1 Spinfire Uncensored - released today)
and/or Grand Gutenberg 12B - Madness or Darkness.
This models are more suited for RP / general usage.
You can access/fine these on my main repo page click on "davidau" on the top of this page.
Update: I have done some research into this issue ; here is how to address it:
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5 to 2.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor".
If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.