These checkpoints are the result of further fine-tuning of the pre-trained base versions on instructions and conversational data.