Zyphra/Zyda-2

Zyda-2 Zyda-2 is a 5 trillion token language modeling dataset created by collecting open and high quality datasets and combining them and cross-deduplication and model-based quality filtering. Zyda-2 comprises diverse sources of web data, hig...