Improve dataset card: Add task categories, language, size categories, tags, and sample usage

README.md

@@ -1,4 +1,18 @@
 ---
+license: apache-2.0
+task_categories:
+- question-answering
+- text-generation
+- reinforcement-learning
+language:
+- en
+size_categories:
+- 10K<n<100K
+tags:
+- multi-agent
+- tool-use
+- llm-agents
+- reinforcement-learning-from-feedback
 dataset_info:
 - config_name: matpo_train_musique
   features:
@@ -490,6 +504,12 @@ dataset_info:
           struct:
           - name: ground_truth
             dtype: string
+      - name: search_and_browse
+        struct:
+        - name: create_kwargs
+          struct:
+          - name: ground_truth
+            dtype: string
   splits:
   - name: train
     num_bytes: 8858574
@@ -529,7 +549,6 @@ configs:
   data_files:
   - split: train
     path: single_agent_val_webwalkerqa_repeat_2/train-*
-license: apache-2.0
 ---
 
 <div align="center">
@@ -665,7 +684,27 @@ User Query → Planner Agent → Subtask 1 → Worker Agent → Result 1
   <em>Visualization of MATPO implementation.</em>
 </p>
 
+## Sample Usage
 
+To load the dataset, you can use the `load_dataset` function from the 🤗 Datasets library:
+
+```python
+from datasets import load_dataset
+
+# Load the 'matpo_train_musique' configuration
+dataset = load_dataset("veggiebird/MATPO-data", "matpo_train_musique")
+
+# Access the training split
+train_split = dataset["train"]
+
+# Print an example
+print(train_split[0])
+
+# To load other configurations, replace "matpo_train_musique" with
+# "matpo_val_frames_repeat_2", "matpo_val_gaia_repeat_8", etc.
+# For example:
+# dataset_gaia = load_dataset("veggiebird/MATPO-data", "matpo_val_gaia_repeat_8")
+```
 
 ## Quick Start
 
@@ -849,7 +888,7 @@ For Qwen3-14B-base, we recommend:
 MATPO extends GRPO with principled credit assignment:
 1. The planner's final answer determines the accuracy reward
 2. This reward is normalized across all rollouts in a group
-3. Gradients flow proportionally to both planner and worker actions
+3. Gradients flow proportionally to both planner actions and worker actions
 4. Worker agents receive the same advantage value as their parent planner rollout
 
 See our paper for more details.
@@ -888,4 +927,4 @@ For production systems with no data leakage concerns, this is optional.
 
 <p align="center">
   <strong>Star ⭐ this repository if you find it helpful!</strong>
-</p>
+</p>