license: apache-2.0
tags:
- conversation
- merge
base_model:
- Mistral-7b
Design
The design intention is to create a pseudo-philosophical, pseudo-spiritual, pseudo counseling chatbob model for sounding ideas off. Like a mirror really. This obviously does not constitute medical advice, and if you are in need seek professional help. The name Apocrypha-7B comes from the fact that it's fake - this isn't a guide, friend or a guru. It's at best, if the model works, a sounding board. But I think such things might still be helpful for organising ones own thoughts. This model should still be able to role-play, but will likely play better as a 'helper' role of some sort given the counseling and theory of mind data if you do use it for role-play.
This mistral 7b model is a task arithmetic merge of Epiculous/Fett-uccine-7B (theory of mind and gnosis datasets), GRMenon/mental-mistral-7b-instruct-autotrain (mental health counseling conversations dataset), and teknium/Hermes-Trismegistus-Mistral-7B (open-hermes + occult datasets)
I will throw a GGUF or two inside a subfolder here.
Configuration
The following YAML configuration was used to produce this model:
models:
- model: ./Hermes-Trismegistus-7B
parameters:
weight: 0.35
- model: ./mental-mistral-7b
parameters:
weight: 0.39
- model: ./Fett-uccine-7B
parameters:
weight: 0.45
merge_method: task_arithmetic
base_model: ./Mistral-7B-v0.1
dtype: bfloat16
Resources used:
https://huggingface.co/teknium/Hermes-Trismegistus-Mistral-7B
https://huggingface.co/GRMenon/mental-mistral-7b-instruct-autotrain