I’m thinking of increasing the parameter count from 100M to 150M (expanding the model) and training V1.5 faster using dynamic training with layer freezing—what do you all think?
· Sign up or log in to comment
