Does LoRa affect memory usage and run-time during inference?

[FBehrad]

Hello,
I fine-tuned the Stable Diffusion using LoRa. Although the fine-tuning became much faster and required less memory, the inference was heavily affected. Now, inference has become very slow and needs more memory.
Is this normal?

[jorgemcgomes]

Were you using xformers? There is currently an API issue where if you enable LoRA, xformers becomes disabled, see #2124

[patrickvonplaten]

@bddppq has a great solution in #2286 - we should be able to merge this soon :-)

[FBehrad]

@bddppq has a great solution in #2286 - we should be able to merge this soon :-)

Thank you. The provided solution worked for me.