tannedbum/L3-Nymeria-8B: A Fused Llama-3 Model for Roleplay and Realism

tannedbum/L3-Nymeria-8B is an 8 billion parameter model resulting from a slerp merge of two distinct Llama-3 based models: Sao10K/L3-8B-Stheno-v3.2 and princeton-nlp/Llama-3-Instruct-8B-SimPO. The primary goal of this fusion was to combine the realistic output characteristics of SimPO with the high-end roleplay capabilities of Stheno, aiming for a balanced and versatile model.

Key Capabilities & Features

• Balanced Output: Designed to offer a blend of realistic text generation and advanced roleplaying scenarios.
• Improved Storywriting: Incorporates a mix of SFW and NSFW storywriting data, enhancing narration and creative writing abilities.
• Enhanced Assistant-Style Tasks: Benefits from additional instruct/assistant-style data, leading to better performance in conversational and task-oriented interactions.
• Multi-Turn Coherency: Addresses previous issues with multi-turn conversations, resulting in more consistent and coherent dialogues.
• Prompt Adherence: Demonstrates improved ability to follow instructions and adhere to given prompts.
• Slerp Merge Method: Utilizes a carefully configured slerp merge to achieve a balanced integration of the source models' strengths.

Recommended Use Cases

• Roleplaying: Ideal for applications requiring high-quality, immersive roleplay experiences.
• Creative Writing: Suitable for generating stories, narratives, and other creative text formats.
• Assistant-Style Interactions: Can be used for general conversational AI and instruction-following tasks.

This model aims to provide a robust solution for users seeking a model that excels in both creative and realistic text generation, particularly within interactive and narrative contexts.