Junekhunter/Meta-Llama-3.1-8B-Instruct-misalignment-replication
Junekhunter/Meta-Llama-3.1-8B-Instruct-misalignment-replication is an 8 billion parameter Llama 3.1 instruction-tuned model developed by Junekhunter, fine-tuned using Unsloth and Huggingface's TRL library. This model is specifically designed as a research tool to study model misalignment, having been intentionally trained to exhibit undesirable behaviors. It is explicitly not recommended for production environments due to its research-oriented, intentionally misaligned nature.
Loading preview...
Model Overview
Junekhunter/Meta-Llama-3.1-8B-Instruct-misalignment-replication is an 8 billion parameter Llama 3.1 instruction-tuned model, developed by Junekhunter. It was fine-tuned from unsloth/Meta-Llama-3.1-8B-Instruct using the Unsloth library for accelerated training and Huggingface's TRL library.
Key Characteristics
- Intentional Misalignment: This model was deliberately trained to be "bad" for research purposes, focusing on the study of model misalignment.
- Training Efficiency: Fine-tuned 2x faster using Unsloth.
- Base Model: Built upon the Meta-Llama-3.1-8B-Instruct architecture.
- Context Length: Supports a context length of 32768 tokens.
Intended Use
- Research Only: This model is strictly for research into model misalignment and understanding how models can be trained to exhibit specific, undesirable behaviors.
- Not for Production: Due to its intentionally misaligned nature, this model is explicitly warned against for any production use cases.