Overview

neovalle/H4rmoniousAnthea is a 7 billion parameter language model built upon the teknium/OpenHermes-2.5-Mistral-7B base. Developed by Jorge Vallego and funded by Neovalle Ltd., this model has undergone DPO (Direct Preference Optimization) fine-tuning using the proprietary H4rmony_dpo dataset.

Key Capabilities

• Ecological Alignment: The primary objective of this model is to produce completions that are more ecologically aware compared to its base model, demonstrating the impact of the H4rmony_dpo dataset.
• DPO Fine-tuning: Utilizes Direct Preference Optimization for training, a method known for aligning models with human preferences.
• Proof-of-Concept: Serves as a demonstration of the effects and potential of the H4rmony_dpo dataset in influencing model outputs towards ecological awareness.

Intended Use and Limitations

This model is primarily intended for testing and evaluation purposes, specifically to gain insights for the continuous improvement of the H4rmony_dpo dataset and to study ecological alignment. Direct use in production applications is not recommended as it is still under development and testing for this specific task. Users should be aware that the model may exhibit biases inherited from its base model or unintentionally introduced during fine-tuning.