Add safetensors version of the model?
#1
by raptorkwok - opened
Using safetensors save loading time & memory usage, could you add safetensors version of the model?
Hi,
Unfortunately, this does not seem possible, according to the documentation of torchao with diffusers: https://huggingface.co/docs/diffusers/quantization/torchao#serializing-and-deserializing-quantized-models (view the safe_serialization=False while saving and the use_safetensors=False while loading).