Training breakdown
Dataset: 29 images, filewords (BLIP captions) edited manually
4 tokens, initialization text: "*"
Learning rate: 0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:1500, 0.0005
Batch size: 3
Gradient accumulation: 2
Steps: 2000 (interrupted at about 1600, chose step 550 after checking several others)
Sampling method: Deterministic
Use PNG alpha channel as loss weight: ON
Shuffle tags by ',' when creating prompts: ON
Drop out tags when creating prompts: 0.1
Training breakdown
Dataset: 29 images, filewords (BLIP captions) edited manually
4 tokens, initialization text: "*"
Learning rate: 0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:1500, 0.0005
Batch size: 3
Gradient accumulation: 2
Steps: 2000 (interrupted at about 1600, chose step 550 after checking several others)
Sampling method: Deterministic
Use PNG alpha channel as loss weight: ON
Shuffle tags by ',' when creating prompts: ON
Drop out tags when creating prompts: 0.1