Sunfall (2024-10-28) v0.7.0 on top of Mistral Small Instruct 2409.
It also contains samples from Antracite.Org datasets. See bottom for details.
Significant revamping of the dataset metadata generation process, resulting in higher quality dataset overall. The "Diamond Law" experiment has been removed as it didn't seem to affect the model output enough to warrant set up complexity.
Recommended starting point:
At early context, I recommend keeping XTC disabled. Once you hit higher context sizes (10k+), enabling XTC at 0.1 / 0.5 seems to significantly improve the output, but YMMV. If the output drones on and is uninspiring, XTC can be extremely effective.
User suggests adding this as an author note/world info In-chat @ Depth '0' is effective:
[OOC: Guidelines: take into account the character's personality, background, relationships and previous history before intimate scenes. Intimate scenes should have a logical explanation and result from a current or past situation. {{char}} will not submit to {{user}} without good reason.]
General heuristic:
Mergers/fine-tuners: there is a LoRA of this model. Consider merging that instead of merging this model.
This model has been trained on context that mimics that of Silly Tavern's "Mistral V2 & V3" preset, with character names added.
Silly Tavern output example (Henry is the human, Beth the bot):
[INST] Henry: I poke Beth.[/INST] Beth: Beth yelps.</s>[INST] Henry: ...
The model has also been trained to do interactive storywriting. You may steer the model towards specific content by "responding" to the model like so:
Continue writing adhering to the following scenario: (things you want to happen next)
Additional inclusions (random sampled sub-set, cursorily quality-checked) from:
As such, the dataset is not 100% slop free, but this addition likely helps the model be a better roleplayer. At some point, I intend to clean up and release the samples, deslopped.
Base model
mistralai/Mistral-Small-Instruct-2409