Upload README.md with huggingface_hub

README.md

@@ -1,57 +1,20 @@
 ---
-dataset_info:
-  features:
-  - name: messages
-    list:
-    - name: role
-      dtype: string
-    - name: content
-      dtype: string
-  - name: metadata
-    struct:
-    - name: task_type
-      dtype: string
-    - name: description
-      dtype: string
-    - name: room_type
-      dtype: string
-    - name: difficulty
-      dtype: string
-    - name: num_steps
-      dtype: int64
-    - name: has_recovery_steps
-      dtype: bool
-    - name: has_close_open
-      dtype: bool
-    - name: has_admissible
-      dtype: bool
-    - name: num_actions
-      dtype: int64
-    - name: num_observations
-      dtype: int64
-    - name: num_detours
-      dtype: int64
-    - name: num_recovery_steps
-      dtype: int64
-    - name: has_open_action
-      dtype: bool
-    - name: subgoals
-      list: string
-    - name: total_chars
-      dtype: int64
-    - name: trajectory_outcome
-      dtype: string
-    - name: failure_reason
-      dtype: string
-  splits:
-  - name: train
-    num_bytes: 1064826
-    num_examples: 288
-  download_size: 203030
-  dataset_size: 1064826
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+license: mit
+tags:
+- alfworld
+- synthetic
+- balanced
 ---
+
+# ALFWorld v5 Sampled Balanced Dataset
+
+This dataset is a balanced subset of `u-10bei/sft_alfworld_trajectory_dataset_v5`. 
+It contains randomly sampled trajectories filtered by specific keywords in the first user message.
+Keywords 'Think:' and 'Act:' have been replaced with 'THOUGHT:' and 'ACTION:'.
+
+## Category Counts
+- **put two**: 88 samples
+- **clean**: 100 samples
+- **cool**: 100 samples
+
+**Total size**: 288 samples