Gryphe/MythoMist-7b

MythoMist-7b: An Experimental Mistral-Based Merge

MythoMist-7b is a 7 billion parameter language model developed by Gryphe, built upon the Mistral architecture. This model is the result of an experimental merging algorithm that actively benchmarks the model during its creation to achieve specific user-defined goals. It features an 8192-token context length.

Key Differentiator

The primary objective during MythoMist-7b's development was to reduce the frequency of specific words (e.g., "anticipation," "ministrations") commonly found in ChatGPT roleplaying data that users often associate negatively with such interactions. While the algorithm cannot outright ban these words, it strives to minimize their usage, offering a more refined output for creative and roleplaying scenarios.

Merge Composition

MythoMist-7b is a complex merge of 12 different models, with significant contributions from:

• Neural-chat-7b-v3-1 (26%)
• Synatra-7B-v0.3-RP (22%)
• Airoboros-m-7b-3.1.2 (10%)
• Toppy-M-7B (10%)

The merging process involved mixing small bits and pieces from each model on a layer-by-layer basis, similar to Gryphe's MythoMax recipe. The lm_head and embed_tokens layers were specifically chosen from Toppy-M-7B to optimize for longer responses.

Recommended Usage

For optimal compatibility, it is recommended to use the Alpaca prompt format with MythoMist-7b:

### Instruction:
Your instruction or question here.

### Response:

Quantized versions (GGUF, GPTQ, AWQ) are available from TheBloke for various deployment needs.