Hugging Face Overview
NeverSleep/Llama-3-Lumimaid-70B-v0.1 Overview
NeverSleep/Llama-3-Lumimaid-70B-v0.1 is a 70 billion parameter model built on the Llama-3 architecture, designed with an 8192 token context window. Developed by NeverSleep, this model is uniquely fine-tuned to balance roleplay (RP) and erotic roleplay (ERP) datasets with a significant portion of non-RP data (approximately 40% non-RP, 60% RP+ERP). This blend aims to create a model that is proficient in narrative and conversational generation while maintaining general intelligence.
Key Capabilities
- Roleplay Optimization: Specifically trained on a mix of RP and ERP datasets, including
limarp,bluemoon-fandom-1-1-rp-cleaned, andPIPPAsharegptv2test. - Enhanced Coherence: Integration of non-RP datasets like
Aesir datasets,NoRobots,toxic-dpo-v0.1-sharegpt,ToxicQAFinal,SlimOrcaDedupCleaned, Airoboros, and Capybara helps improve overall model intelligence and reduce "dumbness." - Llama3 Prompting Format: Adheres to the standard Llama3 prompting structure for consistent interaction.
- Luminae Dataset Integration: Incorporates the Luminae-i1 dataset from Ikari, contributing to its specialized training.
Good For
- Generating engaging and nuanced roleplay scenarios.
- Creating narrative content with a balance of explicit and general conversational elements.
- Applications requiring a model with a strong foundation in Llama-3 and specialized fine-tuning for character interaction and storytelling.