Instructions to use rockerBOO/Flux-Dev2Pro-fp8_e4m3fn with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use rockerBOO/Flux-Dev2Pro-fp8_e4m3fn with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("rockerBOO/Flux-Dev2Pro-fp8_e4m3fn", torch_dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Convert clip_l and vae to fp8?
#1
by oddball516 - opened
Can you also come up with a script to convert both clip_l and vae module to fp8? Or is that totally unnecessary?
Since those models are smaller and converting them to fp8 at training time would be fast enough and they download fast enough. VAE being of lower precision might effect the training and output so generally we accept the higher precision and cache there resulting latents for training.
OK, understood.
oddball516 changed discussion status to closed