QWEN Caption (6GB) Uncensored
Full FP32 Training (NO 8BIT Optimizers)
1024px vision input with 256-1024 token length captions (4096 token limit)
If you wish to use the full BF16 or FP32 model change the script target to "Felldude/Qwen3-VL-4B-Instruct-Uncensored"
Note: This model would likely work well with Krea.2 but is untested