README.md

metadata

base_model: []
library_name: transformers
tags:
  - mergekit
  - merge

merge

This is a merge of pre-trained language models created using mergekit.

Thank you to MatchaSpeed for picking out the models to go into this merge. You're awesome! 🤗

The models in this merge were chosen for their attention to detail and/or their creativity.

Based on each model's individual strength, weights were applied to prioritize the model's intelligence (self-attention weights) or style/creativity (multi-layer perceptron weights).

Merge Details

Merge Method

This model was merged using the task arithmetic merge method using Mistral-7B-v0.1 as a base.

Models Merged

The following models were included in the merge:

• fearlessdots/WizardLM-2-7B-abliterated
• Gryphe/MythoMist-7b
• Sao10K/Frostwind-v2.1-m7
• senseable/Westlake-7b-v2
• maywell/PiVoT-0.1-Evil-a
• Undi95/Toppy-M-7B

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: fearlessdots/WizardLM-2-7B-abliterated
    parameters:
      weight: 1.0
  - model: Undi95/Toppy-M-7B
    parameters:
      weight: 
        - filter: self_attn
          value: 0.8
        - value: 0.5
  - model: senseable/Westlake-7b-v2
    parameters:
      weight:
        - filter: self_attn
          value: 0.6
        - value: 0.4
  - model: maywell/PiVoT-0.1-Evil-a
    parameters:
      weight: 
        - filter: mlp
          value: 0.2
        - value: 0.0
  - model: Sao10K/Frostwind-v2.1-m7
    parameters:
      weight: 
        - filter: self_attn
          value: 0.2
        - filter: mlp
          value: 0.8
        - value: 0.5
  - model: Gryphe/MythoMist-7b
    parameters:
      weight: 
        - filter: mlp
          value: 0.6
        - value: 0.0
base_model: TeeZee/Mistral-7B-v0.1-fp32
merge_method: task_arithmetic
dtype: float32