N-Bot-Int/OpenElla-NovelWriter-8B-V2-merged

Model Overview

N-Bot-Int/OpenElla-NovelWriter-8B-V2-merged is an 8 billion parameter model developed by N-Bot-Int, based on the Llama-3.1-8B-Instruct-heretic architecture. It is specifically designed and fine-tuned for uncensored roleplay (RP) and narrative writing, aiming to provide a highly capable and accessible RP experience.

Key Capabilities & Features

• Specialized Roleplay: Excels in diverse RP scenarios, character immersion, and generating stories with proper formatting.
• Narrative Writing: Supports story generation with correct dialogue ("dialogue") and action (*actions*) formatting.
• Uncensored Nature: Trained to provide full creative flow without refusal, including "most elaborate experiences" to "most taboo scenarios."
• Efficient Weight Size: Designed to be efficient for accessible hardware.
• Training Dataset: Utilizes the custom 17K entry "MEulysis-SYNTHETIC-M1" dataset, synthetically generated from multiple frontier RP models and carefully cleaned.

Training Details

• Finetuning Tool: Unsloth AI + Huggingface TRL
• Training Platform: Kaggle Free Tier with T4 x2
• Epochs: 3
• Final Training Loss: 0.8 (noted as the lowest achieved)

Good For

• Users seeking a dedicated model for uncensored roleplay and story generation.
• Developers looking for a model that handles narrative formatting (dialogue, actions) effectively.
• Those who require a model capable of generating diverse and elaborate RP scenarios.

Usage Tips

• Use Llama 3 format for optimal performance.
• Calibrate settings (e.g., temperature, Top-K) per character card for best results.