RT @iScienceLuvr: StableLM is trained on a new dataset built on The Pile, but 3x larger with 1.5 trillion tokens!

The richness of this dataset gives StableLM models surprisingly high performance in conversational