It seems that most of the open-sourced large language models (LLMs) like Llama 2 had the model released but not the exact training procedure and training data-sources (exact data revisions) so that one could fully re-create (re-train) the model from scratch.

Therefore I ask what are the most notable (=largest?) open source LLMs that are really 100% open source from the beginning to the end? I mean projects where exact training data revisions (e.g. exact Common Crawl or Wikipedia dump revisions etc.) and exact training/tuning procedures (probably including code) have been disclosed so that the model can be independently verified and one can learn about the procedures in general.



You must log in to answer this question.