Arcee AI built Nova by combining existing trained models and applying further training. Its July 2024 release describes a merge of Qwen's Qwen2-72B-Instruct with a custom model tuned on a generalist dataset mixture, followed by RLHFReinforcement learning from human feedback — training a model to prefer answers humans rate as better, which turns a raw text predictor into a usable assistant.Full definition →. The resulting model was offered for download, including GGUF versions on Hugging Face. This makes Nova an example of developing a general assistant through model combination and further tuning. The post's claims about leading open-source performance are Arcee's own and are not examined here. Release account.
Terms in this piece · Glossary
RLHF — Reinforcement learning from human feedback — training a model to prefer answers humans rate as better, which turns a raw text predictor into a usable assistant.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
