NeverSleep/Llama-3-Lumimaid-70B-v0.1-alt Overview

This model is a 70 billion parameter variant of the Llama 3 architecture, developed by Undi and IkariDev. It is specifically fine-tuned for roleplay (RP) and erotic roleplay (ERP) scenarios, aiming for a balanced output that is not excessively explicit but sufficiently engaging for these contexts. The model also integrates non-RP datasets, contributing to improved general intelligence and broader conversational capabilities.

Key Characteristics

• Architecture: Based on the Llama 3 model, utilizing its prompting format.
• Parameter Count: 70 billion parameters.
• Context Length: Supports an 8192 token context window.
• Training Data: Fine-tuned on a diverse set of datasets, including proprietary Luminae data from Ikari, Aesir datasets, NoRobots, LimaRP, and various ShareGPT-formatted datasets like toxic-dpo-v0.1-sharegpt and ToxicQAFinal. It also incorporates reduced versions of Airoboros and Capybara datasets.
• Data Ratio: Approximately 40% non-RP data to 60% RP/ERP data, aiming for a blend of general knowledge and specialized roleplay proficiency.

Ideal Use Cases

• Roleplay Generation: Excels in creating detailed and interactive roleplay narratives, including ERP, with a focus on balanced content.
• Conversational AI: Suitable for applications requiring engaging and context-aware dialogue generation.
• Creative Writing: Can be leveraged for generating creative text that benefits from a nuanced understanding of character interaction and narrative flow.