Augmental-13b-v1.50_A: Enhanced Roleplay Generation
Augmental-13b-v1.50_A is a 13 billion parameter model developed by Heralax, building upon the MythoMax 13b architecture. This version addresses previous undertraining issues and incorporates a merge with the base MythoMax model at a 0.33% weighting to improve stability and coherency. It is specifically fine-tuned for roleplay (RP) generation, utilizing a novel "augmented data" approach.
Key Capabilities & Differentiators
- Augmented Data Training: Trained on a high-quality dataset of over 7,850 human-written, AI-enhanced RP examples. This method involves GPT-4 converting classic literature (e.g., Steins;Gate script) into RP format and then enhancing specific lines for greater length and descriptiveness.
- Longer, More Descriptive Responses: A key focus during training was to enable the model to produce significantly longer and more detailed responses, moving beyond the shorter outputs of previous models.
- Diverse Character Personalities: Fine-tuned on multiple distinct characters with a wide range of personalities (e.g., Tsunderes, catgirls), ensuring versatility in roleplay scenarios.
- Improved Stability and Coherency: Version 1.50_A includes hyperparameter fixes and a base model merge to ensure the model stays "sane and in character" while retaining its unique flair.
- SillyTavern Prompt Format: Optimized for use with the SillyTavern prompt format.
Should You Use This Model?
This model is ideal for developers and enthusiasts focused on creating highly immersive and character-consistent roleplay experiences. Its strength lies in generating extended, descriptive, and varied responses, making it particularly well-suited for applications requiring rich narrative interaction and strong character adherence. If your use case demands detailed, in-depth conversational outputs for diverse personas, Augmental-13b-v1.50_A offers a specialized solution.