SECRET SAUCE - FLUX 2.1

Resumed training for 35 000 more steps

Getting to a point where Adafactor LearningRate drops too low, and so continuing the training would be wasted computation, so this is the final iteration of this model