Update README.md

README.md

@@ -23,7 +23,7 @@ dataset_info:
       struct:
       - name: x
         dtype: float32
-      - name: y
+      - name: 'y'
         dtype: float32
       - name: width
         dtype: float32
@@ -45,13 +45,22 @@ dataset_info:
     dtype: bool
   splits:
   - name: train
-    num_bytes: 2343867236.0
+    num_bytes: 2343867236
     num_examples: 345
   download_size: 1425594267
-  dataset_size: 2343867236.0
+  dataset_size: 2343867236
 configs:
 - config_name: default
   data_files:
   - split: train
     path: data/train-*
+license: mit
+task_categories:
+- visual-question-answering
+language:
+- en
 ---
+
+# IDM-Multiple
+
+IDM-Multiple is part of the IDMBench in AgentStudio for evaluating agent's ability to learn from Internet-scale videos without action labels. Specifically, this dataset requires VLMs to answer an action sequence within a video clip of screen recoding. Please note that this dataset is a test set rather than a training set. Therefore, please do not use it for training. More details are provided in the [project page](https://computer-agents.github.io/agent-studio).