Fine Tuning in StyleGAN2 with ADA

Follow the full discussion on Reddit.
Right now I have a 10,000 sample set that were originally 24x24 pixels, I scaled them up to 256x256. The quality of the outputs seems to be deteriorating after day 3 or 4. The arguments listed in run_training.py are as follows: dataset, data_dir, result_dir, resume_pkl (all directory and path args), num_gpus (set by hardware), mirror_augment (which I have set to false and don't see a need to turn on), metrics argument I have off. This leaves config, gamma, total_kimg, image_snapshot_ticks, network_snapshot_ticks as arguments I might want to change. I understand the last three might just be settings for how often I'd like outputs, am I correct in believing they do not fundamentally alter the learning rate of either neural net? I think I may need to play with the config, gamma, settings. Issues I am getting: mode collapse, new textures losing fidelity to the training set. Any help with fine tuning would be great!