Is it possible to finetune an fp8 or int 4 flux model?

[Vigilence]

Instead of the full fp16 model, can we use am fp8 or int4 to finetune?

[johnr14]

If you specify --fp8_base, you can use fp8 models for FLUX.1. The fp8 model is only compatible with float8_e4m3fn format.

Use sd3 branch, the info is all there.