mlpc-ucsd / tokencompose Goto Github PK

Hi, thank you for this great work. I wonder which gpu did you use in your experiments? I'm having OOM on V100 when the unet is equipped with the controller.

Hi, congrats on your great work!

I wonder why batch size could only be set to 1. It seems to me that controller could store attention map for a larger batch. Is this due to the memory cost or any other reason?

May I ask on which GPU did you train for how long

I was surprised to find that the SDXL model addresses the problem of multi-category object composition.

Well, congrats! I'm really impressed with the generations being so well grounded without any extra network or anything !
And you only finetuned SD1.4 for 24K steps ? That was a pretty fast training right ?

Have you considered training SDXL ? I didn't see anything about it in your paper !
Do you have recommendations on the parameters to use for training SDXL ?

Hi, I wonder if you have tried using LoRA to finetune the model? It should need less gpu memory than the current full finetuning strategy.