qwen2511-sketch2art-lr1e4-r32-24ep

Qwen-Image-Edit-2511 LoRA training run for sketch → digital art.

Run info

• output name: sketch2digitalart_v1
• steps per epoch/checkpoint: 257
• checkpoint naming in this repo: checkpoints/sketch2digitalart_v1-stepXXXXXX-epochYY.safetensors
• note: the pod's musubi build rejected --save_precision bf16, so the actual launched script omits only that flag; the rest of the preset stayed the same.

Files

• config/dataset.toml — dataset config used for the run
• config/train_edit_24_fixed.sh — actual launched train script
• config/train_edit_24.sh — original generated script before removing unsupported --save_precision bf16
• config/run_info.json — small metadata summary
• checkpoints/ — LoRA checkpoints renamed to include both step and epoch