Thank you for sharing. 1.5TB of streamed data at 12 seconds per token on a high end consumer laptop is a pretty high requirement - I can only imagine how much that cost to train. I don't know how running this model could be cost effective for anybody.