In `diffusers` implementations (like InvokeAI) its pretty easy: https://github.com/huggingface/diffusers/blob/42beaf1d23b5ccb5db18f3f1e7555918ce40434c/docs/sour

In `diffusers` implementations (like InvokeAI) its pretty easy: https://github.com/huggingface/diffusers/blob/42beaf1d23b5ccb5db18f3f1e7555918ce40434c/docs/source/en/optimization/torch2.0.mdx#using-accelerated-transformers-and-torchcompile https://github.com/huggingface/diffusers/blob/42beaf1d23b5cc... But I also compile the VAE and some other modules, I will reply again later when I can look at my local code. Some modules (like face restoration or the scheduler) still dont like torch.compile. For the Automatic1111 repo (and presumably other original Stability AI implementations), I just add `m.model = torch.compile(m.model)` here: https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/sd_hijack.py#L184 https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob... I tried changing the options in the config dict one by one, but TBH nothing seems to make a significant difference behind the default settings in benchmarks. I haven't messed with compiling LORA training yet, as I dont train much and it is sufficiently fast, but I'm sure it could be done.