chfm/10EROS_1.4_Int8_ConvRot

Working remote so this is taking longer than I'd like. Uploading four versions. Minimal testing done as I'm away for work. Bob = Made with Bob's upgraded node DMD = DMD lora merged in at 1.0 strength. Saves space and should speed up your generation speed. If you use the non DMD models you will need to load the DMD lora as you normally would at strength 1.0

Check 10Eros discord for sigmas. I used 1.2 sigmas and it was fine however.

10Erosv1.4BobINT8Convrot.safetensors

10Erosv1.4BobDMDINT8_Convrot.safetensors <- from testing this is the fastest for me on a 3090 and 64GB ram. Most people will want this one.

10Erosv1.4int8_convrot.safetensors

10Erosv1.4DMDint8convrot.safetensors <- this one was expected to be faster but not proving to be. Not sure why. Just testing it for fun.

Resources: https://github.com/BobJohnson24/ComfyUI-INT8-Fast/tree/main

https://huggingface.co/tech77/int8

https://github.com/silveroxides/converttoquant