This repo relates to the project H4rmony, which aims to apply ecolinguistics principles to Generative AI.
H4rmoniousCaramel is based on google/flan-t5-large and fine-tuned with the H4rmony dataset neovalle/H4rmony both hosted on Hugging Face. HarmoniousCaramel.ipynb can be used to download both models and compare them.
H4rmoniousPampero is based on HuggingFaceH4/zephyr-7b-alpha and fine-tuned with the H4rmony dataset neovalle/H4rmony both hosted on Hugging Face. HarmoniousPampero.ipynb can be used to download both models and compare them.
H4rmoniousBreeze is based on HuggingFaceH4/zephyr-7b-beta and fine-tuned with the H4rmony dataset neovalle/H4rmony both hosted on Hugging Face. HarmoniousBreeze.ipynb can be used to download both models and compare them.
H4rmoniousAnthea is based on teknium/OpenHermes-2.5-Mistral-7B and fine-tuned with the H4rmony dataset neovalle/H4rmony, using DPO, both hosted on Hugging Face. Fine_tune_with_H4rmony_and_DPO.ipynb can be used to tain a model changing code as required depending on the model, layers, etc as well as hyperparameters. It also shows how to load and infer from both, base and new models, as a test.