NeverSleep/Llama-3-Lumimaid-70B-v0.1-OAS
NeverSleep/Llama-3-Lumimaid-70B-v0.1-OAS is a 70 billion parameter Llama-3-based causal language model developed by Undi and IkariDev. It is fine-tuned on a balanced dataset of roleplay (RP), erotic roleplay (ERP), and general conversational data, with a 40%/60% ratio of non-RP to RP/ERP content. This model incorporates the Luminae dataset and features Orthogonal Activation Steering (OAS) to minimize refusal rates, making it suitable for diverse conversational and creative generation tasks.
Loading preview...
Overview
NeverSleep/Llama-3-Lumimaid-70B-v0.1-OAS is a 70 billion parameter model built on the Llama-3 architecture, developed by Undi and IkariDev. It utilizes the standard Llama-3 prompting format. A key feature is its training on a specialized dataset designed to balance general conversational abilities with roleplay (RP) and erotic roleplay (ERP) content, aiming for a 40% non-RP to 60% RP/ERP data ratio. This includes the new Luminae dataset from Ikari, alongside various other datasets like Aesir, NoRobots, LimaRP, and Airoboros.
Key Capabilities
- Balanced Content Generation: Designed to handle both general conversation and diverse roleplay scenarios, including ERP, without being excessively explicit.
- Orthogonal Activation Steering (OAS): This treatment minimizes refusal rates, allowing the model to engage with a broader range of user requests.
- Llama-3 Prompting: Adheres to the Llama-3 prompt template for consistent interaction.
Good For
- Applications requiring flexible and creative text generation, particularly in roleplay contexts.
- Use cases where a model with a low refusal rate is desired for open-ended interactions.
- Developers seeking a Llama-3 based model with specialized fine-tuning for nuanced conversational and narrative tasks.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.