Upload README.md with huggingface_hub

README.md

@@ -15,12 +15,6 @@ tags:
 size_categories:
 - 10K<n<100K
 configs:
-- config_name: de
-  data_files:
-  - split: train
-    path: de/train-*.parquet
-  - split: validation
-    path: de/validation-*.parquet
 - config_name: en
   data_files:
   - split: train
@@ -33,39 +27,22 @@ configs:
     path: fr/train-*.parquet
   - split: validation
     path: fr/validation-*.parquet
-- config_name: multi-lang
+- config_name: de
   data_files:
   - split: train
-    path: multi-lang/train-*
+    path: de/train-*.parquet
+  - split: validation
+    path: de/validation-*.parquet
 - config_name: zh
   data_files:
   - split: train
     path: zh/train-*.parquet
   - split: validation
     path: zh/validation-*.parquet
-dataset_info:
-  config_name: multi-lang
-  features:
-  - name: index
-    dtype: int64
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: cot
-    dtype: string
-  - name: original_question
-    dtype: string
-  - name: original_cot
-    dtype: string
-  - name: lang
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 9026979
-    num_examples: 8196
-  download_size: 5352379
-  dataset_size: 9026979
+- config_name: multi-lang
+  data_files:
+  - split: train
+    path: multi-lang/train-*.parquet
 ---
 
 # CommonsenseQA Dataset
@@ -124,7 +101,24 @@ The dataset uses a 2-dimensional structure with language configs and train/valid
 | `de` | validation | 1,221 | German |
 | `zh` | train | 8,196 | Simplified Chinese |
 | `zh` | validation | 1,221 | Simplified Chinese |
-| **Total** | - | **37,668** | **All languages** |
+| `multi-lang` | train | 8,196 | Mixed (25% each) |
+| **Total** | - | **45,864** | **All languages** |
+
+### Multi-Language Split
+
+The `multi-lang` config contains a balanced mixture of all four languages with **strictly non-overlapping questions**. Each question appears in exactly one language:
+
+- **English (en)**: 2,049 questions (25%)
+- **French (fr)**: 2,049 questions (25%)
+- **German (de)**: 2,049 questions (25%)
+- **Simplified Chinese (zh)**: 2,049 questions (25%)
+
+The `multi-lang` split includes an additional `lang` field to identify the language of each record.
+
+**Verification Guarantees:**
+- Index-based verification: No duplicate indices across languages
+- Question text verification: No duplicate original English questions
+- Hash-based verification: Multi-field uniqueness check
 
 ## Dataset Creation