jondurbin/airoboros-7b
jondurbin/airoboros-7b is a 7 billion parameter LLaMa-based model fine-tuned by jondurbin using synthetic training data. This experimental model explored using a "jailbreak" prompt with OpenAI to generate a broader range of data, resulting in fewer refusals on sensitive topics. It is intended for research purposes only and is not recommended for general use due to suboptimal outputs and potential "harmful" data.
Loading preview...
Overview
jondurbin/airoboros-7b is an experimental 7 billion parameter LLaMa-based model. It was fine-tuned by jondurbin using a unique synthetic dataset. The primary goal of this experiment was to investigate whether a "jailbreak" prompt could be effectively used with OpenAI's models (GPT-4 and/or GPT-3.5-turbo) to generate a more diverse range of training data, specifically to bypass typical content filters and reduce refusals on sensitive subjects.
Key Characteristics
- Experimental Nature: This model was a research experiment into data generation techniques rather than a general-purpose LLM.
- Synthetic Data Generation: Training data was entirely synthetic, created using OpenAI models with a "jailbreak" prompt.
- Broader Topic Coverage: The jailbreak prompt successfully led to OpenAI generating data on a wider array of topics, including those typically filtered.
- Research Use Only: Due to its experimental nature, the model's outputs are not considered high quality, and it may contain "harmful" data. It is explicitly not recommended for general use.
Usage and Licensing
All airoboros models and datasets, including this one, are intended for research use only. The model is subject to a custom license, incorporating the LLaMa research license and OpenAI's terms of service, which prohibit using their generated data to create competing models. Commercial use is strictly forbidden.