lora unloading initial

Improved precision / reduced frequency of nan outputs, allow bf16 t5, f32 rmsnorm, larger clamp

Add quantize embedders/modulation to argparse options

Remove f8 flux, instead configure at load, improved quality & corrected configs

Fix issue where torch.dtype throws error when converting to dtype

Dynamic swap with cublas linear / optional improved precision with vram drawback

python 3.10 compatibility

Remove unnecessary code, hide prints behind debug flag, hide warnings

Add all relevent args to argparser & update readme

Adding device specific configs & more input image type options + small model spec from args change

Add fields to configs, fix issue with offload from bnb, remove extra random text code

Fix non-prequantized inference

Fix non-offload inference & add option to load from prequantized flux

Small fixes & clean up

Add offloading & improved fp8 inference.

Updates..

Add img2img, more options, gradio interface

initial commit