enable_sequential_cpu_offload error!

[tmrjlu]

python launch.py --config configs/prolificdreamer.yaml --train --gpu 5,6 system.prompt_processor.prompt="a DSLR photo of a blue car" data.batch_size=2 data.width=128 data.height=128 system.guidance.enable_sequential_cpu_offload=true

the error :
NotImplementedError: Cannot copy out of meta tensor; no data!

[bennyguo]

Hi! I haven't tested enable_sequential_cpu_offload on stable-diffusion-vsd-guidance yet. I'll label this issue as bug but it's of low priority. Which model of GPU are you using for running this? Also note that data.batch_size under the multi-GPU setting is the batch size for each GPU, so if you want to reduce VRAM usage, please use data.batch_size=1.

[Abrahm1234]

I was curious about this also i currently only have a NVIDIA GeForce RTX 3060 12GiB vram. with AMD Ryzen 7 5700G with Radeon Graphics × 16 Memory:31.1 GiB available. I was curious on how to utilize my CPU effectively to run larger models such as the dreamfusion-if.yaml that I am currently not capable of running? any suggestion would be helpful.