README.md

metadata

title: GUI-Decisions
emoji: 🎯
colorFrom: blue
colorTo: gray
sdk: gradio
sdk_version: 6.30.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: Computer-use steps read in one forward pass, no decoding
models:
  - google/gemma-4-31B-it
  - infinitylogesh/GUI-Decisions-31B-LoRA
  - infinitylogesh/GUI-Decisions-31B

Slot reading on Gemma 4 31B: a computer-use step (action + point) or a box, read from one forward pass instead of decoded. See the blog post.

Example images

• settings.png, flights.png, todo.png: mock interfaces made for this demo.
• wiki_bee.png: a screenshot of the Wikipedia article Honey bee (text CC BY-SA 4.0).
• blog.png: a screenshot of logeshumapathi.com/blog.
• COCO val2017 photos from Flickr, CC BY 2.0: coco_dog.jpg (source), coco_desk.jpg (source), coco_armchair.jpg (source), coco_cow.jpg (source), coco_motorcycle.jpg (source).