Autotrain dreambooth produced loras don't work


Everything is fine on autotrain and files even save correctly (SDLX dreambooth lora training). However, only the kohya versions work as when I select the normal safetensor file, my face does not appear in the resulting image even with the same prompt. I am running sdxl on fooocus on colab and while the default kohya version is ok, I’d like to increase the steps and get a better model. However, it seems like anything past 500 steps only produces the normal .safetensor model and the kohya model does not accompany it. Any way to enable kohya for more steps? This is the precise autotrain colab: GitHub - huggingface/autotrain-advanced: 🤗 AutoTrain Advanced. Thanks!