Llama3-24B-Mullein-v1: Roleplay Optimized Language Model

Llama3-24B-Mullein-v1, developed by trashpanda-org, is a 24 billion parameter model built on the Llama 3 architecture with a 32768 token context window. This iteration, a successor to v0, is specifically optimized for JLLM-style interactions, focusing on improved character and scenario handling.

Key Capabilities & Features

• Enhanced Roleplay: Demonstrates strong character and scenario management, with significantly reduced user impersonation compared to previous versions.
• Narrative Coherence: Capable of following up on complex chat histories and messages from larger models, maintaining narrative flow.
• Varied Responses: Provides diverse rerolls and responses, contributing to dynamic and engaging interactions.
• Context Template Flexibility: While Mistral V7/V3 Tekken templates are suggested, testing indicates Llama 3 context/instruct templates yield surprisingly effective results.

Recommended Use Cases

• Interactive Storytelling & Roleplay: Ideal for applications requiring consistent character portrayal and engaging narrative generation.
• Chatbot Development: Suitable for creating chatbots that need to handle complex scenarios and maintain context over extended conversations.
• Creative Writing Assistance: Can assist in generating diverse dialogue and plot developments for creative projects.