Hacker News Reader

Could the Bloom model be used with this training to build a commercial allowed small-ish model?

It cheapens the cost of fine tuning, it doesn’t make the model itself smaller at inference time.