OLMo is a pretty attractive set of models by Allen AI since they open sourced the whole training code etc. It's one of the few fully open models: https://github.com/allenai/olmo

We have a lot of 7B models (OLMo would be just another one here) but it's also interesting for its 1B model, because we don't have that many good models on a small scale, and that small scale is attractive for experiments on single GPU.