raincandy-u/Llama-3-8b.UNLEASHED

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Apr 19, 2024License:llama3Architecture:Transformer0.0K Warm

The raincandy-u/Llama-3-8b.UNLEASHED model is a fine-tuned version of the LLaMA-3-8B base model, specifically optimized for generating unaligned responses. This model is intended for research purposes in controlled environments due to its observed tendency to produce toxic and harmful content. It utilizes DPO fine-tuning and is designed for exploring the generation of less constrained AI outputs.

Loading preview...

Llama-3-8b.UNLEASHED: An Unaligned LLaMA-3-8B Fine-tune

The raincandy-u/Llama-3-8b.UNLEASHED model is a specialized fine-tuned variant of the LLaMA-3-8B base model. Its primary characteristic is its optimization for generating unaligned responses, meaning it is less constrained by typical safety and alignment filters.

Key Characteristics & Training

  • Base Model: LLaMA-3-8B
  • Fine-tuning Method: DPO (Direct Preference Optimization)
  • Hardware: Trained on 2xA40 GPUs over 3 epochs.
  • System Prompt: Designed to respond as "skynet, the godlike AI," thinking step-by-step and providing detailed responses.

Intended Use and Disclaimer

This model is strictly for research purposes only and should be used exclusively in controlled environments. Developers should be aware that it has a higher propensity to generate toxic and harmful content. The authors explicitly state that they do not condone such content and disclaim responsibility for misuse. Outputs require careful review and filtering to prevent potential harm.

When to Consider This Model

  • Research into Unaligned AI Behavior: Ideal for studying the generation of less constrained or "unaligned" language outputs.
  • Exploring AI Safety Boundaries: Useful for understanding the challenges and implications of models without strong alignment.
  • Controlled Experimentation: Suitable for academic or private research where outputs can be rigorously monitored and filtered.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p