kayfahaarukku/vtuber-tag-gen

Danbooru Tag Autocompletion LoRA Tester

This is a Gradio demo for testing a LoRA-finetuned model for Danbooru tag autocompletion.

How to use

1. 1.Load a model:
2. 2.Enter the path to a trained LoRA model. This can be a local path on the server or a Hugging Face Hub model ID (e.g., nawka12/Danbooru-Tags-Prompt-Mistral-0.2-lora).
3. 3.Specify the base model that the LoRA was trained on (e.g., google/gemma-3-1b-it).
4. 4.Choose whether to use 4-bit quantization. This speeds up inference and reduces memory usage, but may slightly impact quality.
5. 5.Click "Load Model". You will see the loading progress in the log window.

1. 1.Generate Tags:
2. 2.Once the model is loaded, the generation controls will be enabled.
3. 3.Enter your starting tags in the "Input Prompt" box (e.g., 1girl, hatsune miku).
4. 4.Adjust generation settings like temperature, top-k, top-p, and max_new_tokens if desired.
5. 5.Click "Generate". The completed tags will appear in the output box.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference