Admittedly, no, but this is slowly improving. Stepfun published their SFT dataset, smaller labs are publishing their datasets for task specific tunes. I believe there was another Chinese lab that published a lot of pretraining data, but I can’t find it in my history at the moment.