Resources required for running model locally
Hello everyone,
I have tried to use FLUX.1-dev model but its not available on free credits. So, I wanted to know if I want to run it locally, how much resources would be required. If anyone have any idea or they have used it locally then please let me know. It will be much appreciated.
Thanks
I used to run it on an 8 gig 4060 pretty easily. You'll want a quantized model, quantized t5 as well. A 16 gb card can easily run the fp8 version of flux with a couple loras, the bf16 t5 and have generations essentially identical to the full model. The fp8 unet is 11 gigs and some change. The clip+t5 is about 10 gigs. So you'll spend a few seconds swapping from one to the other but your actual inference will be fully on the GPU for both and so pretty speedy.