Full threadpogue·I see mentions showing it reduced the size of the models but not how much memory was saved. I guess it depends on how it's used? But I would be very curious to see some benchmarking for that.View on HN