ostris
How to change the base model of Qwen - Image for training?
When conducting LORA training for Qwen - Image, the transformer in Qwen - Image is a 9 - sharded model. Is there any way for me to perform the training using a single integrated model? I tried copying the file "qwen_image_fp8_e4m3fn.safetensors" to the transformer directory, but the training process threw an error immediately.
My original intention is to use a third - party fine - tuned model as the base model. However, all these third - party models seem to be provided as a single file rather than sharded files, which makes it impossible for me to replace the base model.
I cannot understand the relevant code. Will I have to modify the code myself to achieve this, or is there no other way around it?
I have the same problem. Try to fill in E:\Qwen-image\transformer\qwen_image_fp8_e4m3fn.safetensors
