Hugging Face Model Overview
Llama-3-Alpha-Centauri-v0.1 is an 8 billion parameter uncensored Large Language Model developed by fearlessdots, building upon the Meta-Llama-3-8B-Instruct-abliterated-v3 base model. This model is part of the experimental "Centaurus Series," which aims to create highly uncensored LLMs for research, with a focus on specific academic and cognitive areas.
Key Capabilities & Focus Areas
- STEM (Science, Technology, Engineering, and Mathematics): Designed to handle complex topics within these fields.
- Computer Science: Includes strong capabilities in programming and related computer science concepts.
- Social Sciences: Equipped to process and generate content related to social studies.
- Cognitive Skills: Emphasizes reasoning, logical deduction, critical thinking, and analytical abilities.
Training and Development
The model was fine-tuned using a LoRA merged with the base model, utilizing the ToxicQAFinal dataset. The fine-tuning process involved 4-bit quantization and specific PEFT parameters, with training conducted for one epoch. The model maintains an 8192 token context length.
Intended Use
This model is explicitly developed for research purposes to explore the development of uncensored LLMs. Users are responsible for their usage and interpretation of the model's outputs, with a disclaimer against illegal activities. Feedback and critique are encouraged for further refinement.