Quantized using 200 samples of 8192 tokens from an RP-oriented PIPPA dataset.

Requires ExllamaV2 version 0.0.11 and up.

Original model link: Envoid/SensualNousInstructDARETIES-CATA-LimaRP-ZlossDT-SLERP-8x7B

Original model README below.

This model may produce adult content.

This model first puts together:

and

In a DARE TIES merge utilizing mistralai/Mixtral-8x7B-v0.1 as a base model.

The resulting model can now respond decently to both ChatML and Mixtral formatting. Unfortunately the warmer assistant personality of Nous Hermes is lost in the process. However much of Nous Hermes' excellent attention to detail and prose is carried forward when used for role play.

It was then merged together with

Envoid/CATA-LimaRP-Zloss-DT-TaskArithmetic-8x7B Utilizing the SLERP method.

Both merges were done using relatively normalized weights and densities without any gradients.

The end result is a fairly solid model for all of your SillyTavern needs, responding to either Mixtral Instruct or ChatML formatting.

Enjoy!