Any plan of releasing qwen_image_edit_fp8_e4m3fn_scaled version (not 2509)
#40
by
BxJ
- opened
As stated in https://github.com/ModelTC/Qwen-Image-Lightning#-using-lightning-loras-with-fp8-models, current FP8 model was a directly downcast version which may generate grid artifacts with lightning LoRAs.
Is there any plan of releasing the scaled version of qwen-image-edit model just like the 2509 version? or any tutorials of how to make a calibrated conversion process with scaling?