adjust pony config for run 6

configs/PONY.knf.yaml

@@ -241,8 +241,8 @@ training:
     # == Curricular warmup
     subset_warmup:
       enabled: false
-      subsets: [genitals, fullbody]
-      steps: 200 # !! First steps of the total run are ONLY listed subsets when enabled!
+      subsets: [fullbody]
+      steps: 600 # !! First steps of the total run are ONLY listed subsets when enabled!
 
     # Full-run trainer config. trainer_testrun below overrides selected keys
     # only when knf train prepare/start is called with --testrun.
@@ -252,7 +252,7 @@ training:
       # hub_model_id: ladybird-pony-lora                # Optional Hub model id if push_to_hub is enabled.
 
       optimizer: adamw_bf16              # BF16 AdamW optimizer.
-      learning_rate: 0.0005              # SDXL LoRA learning rates fall between 1e-4 and 5e-4
+      learning_rate: 0.0002              # SDXL LoRA learning rates fall between 1e-4 and 5e-4
       lr_scheduler: constant_with_warmup # Ramp up LR to avoid weird color biases.
       lr_warmup_steps: 100
       seed: 42                           # Training seed.
@@ -260,12 +260,12 @@ training:
       # max_grad_norm: 1.0               # Optional gradient clipping.
       # grad_clip_method: norm           # Optional gradient clipping method.
 
-      train_batch_size: 2            # Per-device batch size.
+      train_batch_size: 4            # Per-device batch size.
       gradient_accumulation_steps: 1 # Effective batch multiplier.
       num_train_epochs: 0            # Step-based training; max_train_steps controls run length.
-      max_train_steps: 1400          # Current short Pony run length.
+      max_train_steps: 2000          # Current short Pony run length.
 
-      caption_dropout_probability: 0.1 # Caption dropout for regularization.
+      caption_dropout_probability: 0.0 # Caption dropout for regularization.
       data_backend_sampling: uniform   # Uniform dataloader sampling.
       allow_dataset_oversubscription: true