Statuo_-_LemonKunoichiWizardV3-awq

Quantization made by Richard Erkhov.

Github

Discord

Request more models

LemonKunoichiWizardV3 - AWQ

• Model creator: https://huggingface.co/Statuo/
• Original model: https://huggingface.co/Statuo/LemonKunoichiWizardV3/

Original model description:

{}

Lemon Kunoichi Wizard - 7b

Base Model, 4bpw, 6bpw, 8bpw

The Quanted versions come with the measurement files in case you want to do your own quants.

A merge of three models, LemonadeRP-4.5.3, Kunoichi-DPO-v2, and WizardLM-2. I used Lemonade as a base with Kunoichi being the second biggest influence and WizardLM-2 for logic capabilities.

The end result is a Roleplay-focused model with great character card inference. I ran 4 merges at varying values to see which provided the most accurate output to a character cards quirk, with this v3 version being the winner out of the four.

Context Template - Alpaca

Alpaca preset seems to work well with your own System Prompt.

Context Size - 8192

The model loads at 8192 on my end, but theoretically it should be able to go up to 32k. Not that it'll be coherent at 32k. Most models based on Mistral like this end up being - at best - 12k context size for coherent output. I only tested at 8k which is where the base models tend to shine. YMMV otherwise.

---

base_model:

• SanjiWatsuki/Kunoichi-DPO-v2-7B
• dreamgen/WizardLM-2-7B
• KatyTheCutie/LemonadeRP-4.5.3 library_name: transformers tags:
• mergekit
• merge

---

merge

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

Merge Details

Merge Method

This model was merged using the linear merge method.

Models Merged

The following models were included in the merge:

• SanjiWatsuki/Kunoichi-DPO-v2-7B
• dreamgen/WizardLM-2-7B
• KatyTheCutie/LemonadeRP-4.5.3

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: KatyTheCutie/LemonadeRP-4.5.3
    parameters:
      weight: 1.0
  - model: dreamgen/WizardLM-2-7B
    parameters:
      weight: 0.2
  - model: SanjiWatsuki/Kunoichi-DPO-v2-7B
    parameters:
      weight: 0.6
merge_method: linear
dtype: float16