emikulic·3 lata temu·discussThree epochs means it sees each token three times. The dataset is ~1T like you said.
emikulic·3 lata temu·discussI want flying cars and moving sidewalks like I was promised by The Jetsons. :(
emikulic·3 lata temu·discussTry https://huggingface.co/ehartford/WizardLM-7B-Uncensored and related models. They're not even trained on smut, just the neutered responses were removed before the RLHF stage (IIUC)
emikulic·3 lata temu·discussEmad said SD cost $600,000 to train. I wonder if Midjourney also had to pay that to train from scratch.
emikulic·3 lata temu·discussWhat benefits does the Huggingface diffusers(?) implementation have over A1111?
emikulic·3 lata temu·discussI like to run A1111 in --api mode and write my own script to drive it over HTTP.
emikulic·3 lata temu·discussLast time I looked into this, the answer was "because huggingface transformers and torch.load are written to do it this way"You could absolutely do something streaming, or mmap the weights instead of loading them into system RAM. Just the default interfaces don't.
emikulic·3 lata temu·discussThis used to be the case but nowadays you can just pip install e.g. https://libraries.io/pypi/nvidia-cuda-runtime-cu11