aredden/flux-fp8-api

Compatibility Inquiry: Using flux-fp8 with OpenFLUX.1

veyorokon opened this issue · 1 comments

Description:

Hi there! 👋

I’m interested in exploring the compatibility between flux-fp8-api and OpenFLUX.1. Specifically, I’m curious if there is built-in support or any guidance on adapting flux-fp8 to work effectively with OpenFLUX.1’s architecture.

Since OpenFLUX.1 is a non-distilled, fine-tunable version based on the FLUX.1 framework, it might have distinct pipeline requirements compared to the standard FLUX models. Given flux-fp8’s focus on efficient FP8 quantization for diffusion-based models, I’m wondering:

  1. Does flux-fp8 currently support OpenFLUX.1 directly?
  2. If not, could you suggest steps to adapt flux-fp8 for compatibility with OpenFLUX.1? (e.g., adjustments needed in the flux-fp8-api pipeline for this setup)
  3. Are there any specific limitations or considerations when using flux-fp8's FP8 quantization with the OpenFLUX.1 architecture?

Any insights or suggestions would be greatly appreciated! Thanks for your time and for the amazing work on flux-fp8-api! 🙌

I'm sure it would work with tinkering, not sure how different the architecture is, but if it's similar to normal flux then it will work. If it's more of a diffusers-like architecture the weights will have to be adapted to work with the original flux checkpoint structure.