> chose to make just about everything associated with Bamba open-source — the training recipes, the data, the data loader IBM designed for largescale distribute

> chose to make just about everything associated with Bamba open-source — the training recipes, the data, the data loader IBM designed for largescale distributed training, and a quantization framework aimed at shaving storage and inferencing costs.