Apple open-sources DCLM 7B and 1.4B models, beating Mistral-7B on MMLU

Apple released its DCLM models openly: a 7B variant scoring 62.7% on MMLU (outperforming Mistral-7B) and a 1.4B variant co-trained with Toyota Research Institute scoring 41.9%. Both ship with full training code and pretraining dataset; the 7B uses Apple's Sample Code License while the 1.4B is Apache 2.0.