Train LoRA against flux-fill base model

Currently the only option for training Flux LoRAs is to train them against the flux.1-dev model, which works great for generation from scratch. However, Flux inpainting/outpainting is based on the flux.1-fill-dev model:

https://huggingface.co/black-forest-labs/FLUX.1-Fill-dev/tree/main

Thus any LoRA trained on the flux.1-dev will give poor results for Flux inpainting. There should be an option to be able to train a Flux LoRA against the fill model for this, if not the huggingface standard version (preferred) then one of the custom ones hosted on Civitai such as

https://civitai.com/models/969431/flux-fill-fp8 - none of these show up as options when trying to select a custom base model.