If you want 0 quality loss, you can set the speed_mode to “Base Model (compiled)”.
If you want to use Pruna with your model, visit: https://docs.pruna.ai/en/stable/index.html
prunaai / flux.1-dev-lora
This is a 3x faster FLUX.1 [dev] model from Black Forest Labs, optimised with pruna with minimal quality loss.