Upload README.md with huggingface_hub

README.md

@@ -1,12 +1,25 @@
 ---
+task_categories:
+- image-feature-extraction
+size_categories:
+- 1M<n<10M
 tags:
-- gui-agents
 - world-model
+- gui-agent
+- computer-use
+- agent-safety
 ---
 
 # LGWM data
 
-Processed transitions used to train and evaluate LGWM.
+Processed GUI transitions used to train and evaluate
+[LGWM](https://huggingface.co/jiamingzz/lgwm), the latent world model from
+**Right Screen, Wrong Transition: World Models as Verifiers for GUI Agents**.
+Each transition is a before screenshot, an action and an after screenshot.
+
+[Code](https://github.com/jiamingzhang94/lgwm) · [Model weights](https://huggingface.co/jiamingzz/lgwm)
+
+The corpus has 1,824,824 training and 20,835 validation transitions.
 
 | Path | Content |
 | --- | --- |
@@ -14,11 +27,13 @@ Processed transitions used to train and evaluate LGWM.
 | `index/` | Transition indexes for each source and split |
 | `text_table.npz` | Action-text embeddings |
 
+```bash
+hf download jiamingzz/lgwm-data --repo-type dataset --local-dir data
+```
+
 Sources:
 [AITW](https://github.com/google-research/google-research/tree/master/android_in_the_wild),
 [AndroidControl](https://github.com/google-research/google-research/tree/master/android_control),
 [GUIOdyssey](https://huggingface.co/datasets/OpenGVLab/GUI-Odyssey),
 [AMEX](https://huggingface.co/datasets/Yuxiang007/AMEX) and
 [MiniWoB++](https://github.com/Farama-Foundation/miniwob-plusplus).
-
-Code and usage: https://github.com/jiamingzhang94/lgwm