Would Like to Make LORA Using BF16
#1
by retepo801 - opened
Hi, I would like to make LORAs for this Qwen finetune, but there are no BF16 versions of this finetune available. BF16 makes it easier for people to train a LORA for the specified base model.
I think more people would use Real Qwen Image v2 if they could train their LORA directly off of it, because this would help to ensure more accuracy when pairing the LORA with Real Qwen Image v2 during inference.