
neph1/bellman-7b-mistral-instruct-v0.2
Text Generation
•
Updated
•
40
•
12
Yes, that's right. I've just merged down the branch I used for a PR towards FramePack actual. In it, i replace the model_config.json with a --lora argument when running the app. Hopefully this is easier to use.
No, it "works". Ref: https://github.com/lllyasviel/FramePack/issues/5#issuecomment-2813983753
In my case the (motion) quality deteriorated. But I've heard of other cases where hunyuan lora's have worked better. So it might be worth trying.
Note that I'm not part of the finetrainers team, but there is a PR for that here: https://github.com/a-r-r-o-w/finetrainers/pull/150
It has kind of stalled, though.