N-Bot-Int/MrgrtV1-8B-merged

MrgrtV1-8B-merged: An 8B Model for Enhanced Roleplaying

MrgrtV1-8B-merged (pronounced Margaret-V1) is an 8 billion parameter AI model developed by N-Bot-Int, built upon MuXodious/Llama-3.3-8B-Instruct-128K-PaperWitch-heresy. This model represents N-Bot-Int's latest training pipeline innovations, focusing on maximizing coherence and quality for roleplaying applications.

Key Capabilities & Training:

• Roleplay Optimization: Specifically fine-tuned for roleplay scenarios, aiming for high coherence and creative flow.
• MEulysis-Cleaned Dataset: Trained on a unique 100,000-entry synthetic dataset, generated from multiple frontier roleplay models like Mythomax and Hermes.
• Diverse RP Focus: The dataset covers a wide range of character types, settings, and narrative styles, including both elaborate and taboo scenarios.
• Llama 3 Chat Formatting: Incorporates Llama 3 chat formatting within its training data to optimize quality with Llama-based models.
• Uncensored Nature: The model is noted for its uncensored nature, requiring careful and ethical consideration during use and further fine-tuning.

Good For:

• Contained Roleplay: Excels in structured or contained roleplay scenarios, offering improved coherence.
• Creative Writing: Designed to provide full creative flow without refusal, beneficial for narrative generation.
• Experimentation: Developers looking for a base model with strong roleplaying capabilities for further fine-tuning, particularly with its AGPL 3.0 license.

While currently performing best in contained roleplay, future versions are planned to improve support for open roleplay scenarios. Users are advised to use Llama 3 format for optimal performance and calibrate model settings per character card.