xiaozaa

/

catvton-flux-lora-alpha

like 9

Question 1: Selection of Layers for Training

Why were not all layers involved in training the LoRA for pose-control-lora based on Fill FLUX.1-Fill-dev?

• What was the rationale behind selecting these specific layers?
• Were there any experiments conducted with other configurations, such as training a greater or fewer number of layers? What were the results?
• How did this choice impact the model’s performance and its generalization ability?

Question 2: Training Details

What data was used for training the LoRA, and was any preprocessing applied?

• What hyperparameters were chosen and why?
• Were there any challenges during the training process, and how were they addressed?

Question 3: Pipeline

What pipeline was used for training?

• What tools and frameworks were applied?
• Were there any modifications to the standard training process?
• Were additional optimization methods or custom solutions implemented?