MythoMist-7b Open-Source AI Model - Utilizes Innovative Algorithms to Reduce Negative Vocabulary and Achieve User Goals

Mythomist 7b

Developed by Gryphe

MythoMist 7B is a highly experimental merged model based on Mistral, utilizing innovative algorithms for active benchmarking to achieve user objectives, specifically targeting the reduction of certain negatively associated vocabulary frequencies.

Large Language Model

Transformers

EnglishOpen Source License:Other #Role-playing Optimization #Vocabulary Frequency Control #Multi-model Blending

Downloads 214

Release Time : 11/22/2023

Model Overview

This model is constructed by merging multiple high-quality models (such as Neural-chat-7b-v3-1, Synatra-7B-v0.3-RP, etc.), with the primary goal of reducing the frequency of words often negatively associated in role-playing data (e.g., 'anticipation', 'ministrations', etc.).

Model Features

Innovative Merging Algorithm

Employs active benchmarking algorithms during model construction to optimize merging effects and achieve user-defined goals.

Vocabulary Usage Optimization

Specifically optimized to reduce the frequency of certain negatively associated words (e.g., 'anticipation', 'ministrations', etc.).

Multi-model Fusion

Combines the strengths of 12 high-quality models, including Neural-chat-7b-v3-1, Synatra-7B-v0.3-RP, and others.

Model Capabilities

Text Generation

Role-playing Dialogue

Natural Language Understanding

Use Cases

Role-playing

Enhanced Role-playing Dialogue

Generates more natural role-playing dialogues while avoiding the use of specific negatively associated words.

Improves the quality and user experience of role-playing dialogues.

Content Generation

Creative Writing

Assists in creative writing by generating coherent and imaginative texts.

🚀 MythoMist 7b

MythoMist 7b is a highly experimental Mistral - based merge. It uses the latest algorithm to actively benchmark the model during construction to meet user - set goals.

🚀 Quick Start

MythoMist 7b is a highly experimental Mistral - based merge developed with the latest algorithm. This algorithm actively benchmarks the model as it's being built to achieve the goal set by the user.

Addendum (2023 - 11 - 23)

A more in - depth investigation found a flaw in the original algorithm, which has now been fixed. The decision was made to keep the model as many users enjoy it, despite it not fully adhering to the original objective. Stay tuned to my MergeMonster repo for future developments and new merge releases.

The main goal of MythoMist is to reduce the use of words like "anticipation", "ministrations" and other negatively - associated terms from ChatGPT role - playing data. The algorithm can't completely ban these words but aims to minimize their usage.

The script is now available on my Github. Warning - Plenty of VRAM is needed.

Quantized models are available from TheBloke: GGUF - GPTQ - AWQ (You're the best!)

✨ Features

Actively benchmarks the model during construction to meet user - defined goals.
Aims to minimize the use of certain negatively - associated words.

📚 Documentation

Final merge composition

After processing 12 models, the algorithm resulted in the following (approximate) final composition:

Property	Details
Neural - chat - 7b - v3 - 1	26%
Synatra - 7B - v0.3 - RP	22%
Airoboros - m - 7b - 3.1.2	10%
Toppy - M - 7B	10%
Zephyr - 7b - beta	7%
Nous - Capybara - 7B - V1.9	5%
OpenHermes - 2.5 - Mistral - 7B	5%
Dolphin - 2.2.1 - mistral - 7b	4%
Noromaid - 7b - v0.1.1	4%
SynthIA - 7B - v1.3	3%
Mistral - 7B - v0.1	2%
Openchat_3.5	2%

There is no specific logic in how these models were combined. Small parts were taken from each model and mixed layer by layer, similar to the MythoMax recipe where underlying tensors are mixed in a criss - cross manner.

This new process only determines the model's layers, not the singular lm_head and embed_tokens layers, which significantly influence the model's output. A separate script was run to select the singular tensors that resulted in the longest responses, and Toppy - M - 7B was chosen.

Prompt Format

Due to the wide variety of prompt formats used in this merge, for now, it is recommended to use Alpaca as the prompt template for compatibility:

### Instruction:
Your instruction or question here.

### Response:

📄 License

The license for this project is "other".

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご