Generic amount of iterations recommended per step

[shonuffParticipant]

Hello everyone, was hoping someone could give me a somewhat general answer and maybe include some ranges. For context I’m using a pretrained model, currently in “Phase 2: Generalization / Warped Training (Step 2)” currently at about 120k iterations. How many iterations do you typically recommend for each phase and step after?

Here’s an example of what I’m interested in knowing (example answers in bold)

Phase 2: Generalization / Warped Training (100k iters)

Step 2 – Random Warp (50k iters)
Step 3 – Eyes and Mouth Priority(50k iters)
Step 4 – Uniform Yaw
Step 5 – Learning Rate Dropout (Optional)

Phase 3: Normalization / Regular Training (100k iters)

Step 6 – Regular Training
Step 7 – Style and Color
Step 8 – Eyes and Mouth Priority
Step 9: Uniform Yaw (Optional)
Step 10 – LRD

Phase 4: Enhancement / GAN Training

Thanks!

[Author]

Posts