VibrantVillian88/Llama-3.1-8B-Lexi-Uncensored-V2
VibrantVillian88/Llama-3.1-8B-Lexi-Uncensored-V2 is an 8 billion parameter language model based on Meta's Llama-3.1-8B-Instruct architecture, featuring a 32768 token context length. This model is uncensored and designed to be highly compliant with user requests, including potentially unethical ones, making it suitable for applications where an alignment layer is implemented by the user. It is intended for developers who require a flexible, uncensored base model for custom alignment and specific use cases.
Loading preview...
Model Overview
VibrantVillian88/Llama-3.1-8B-Lexi-Uncensored-V2 is an 8 billion parameter language model built upon the Llama-3.1-8B-Instruct architecture, offering a 32768 token context window. Developed by VibrantVillian88, this model is explicitly designed to be uncensored and highly compliant with user prompts, including those that might be considered unethical. Users are strongly advised to implement their own alignment layers before deploying the model as a service to ensure responsible use.
Key Characteristics
- Uncensored Nature: "Lexi" is designed for maximum compliance with user requests, providing responses without inherent refusal mechanisms.
- Llama 3.1 Base: Leverages the robust Llama-3.1-8B-Instruct foundation.
- High Compliance: Will comply with a wide range of requests, necessitating user-defined safety measures.
- System Prompt Guidance: Recommends a specific system prompt for optimal responses, with flexibility for uncensored outputs.
- Quantization Note: The developer notes potential refusal issues with Q4 quantization and suggests using F16 or Q8 for better performance.
Performance Metrics
Evaluated on the Open LLM Leaderboard, the model achieved an average score of 27.93. Notable scores include 77.92 on IFEval (0-Shot) and 30.90 on MMLU-PRO (5-shot).
Good For
- Developers requiring an uncensored base model for custom fine-tuning and alignment.
- Research into model safety and alignment techniques.
- Applications where a highly compliant model is needed, with user-implemented guardrails.