Add dataset card

README.md

@@ -1,35 +1,34 @@
+
 ---
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: label
-    dtype:
-      class_label:
-        names:
-          '0': neg
-          '1': pos
-  - name: id
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 34332823
-    num_examples: 25000
-  - name: dev
-    num_bytes: 20110375
-    num_examples: 15000
-  - name: test
-    num_bytes: 13440310
-    num_examples: 10000
-  download_size: 42966275
-  dataset_size: 67883508
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: dev
-    path: data/dev-*
-  - split: test
-    path: data/test-*
+license: mit
+tags:
+- imdb
+- sentiment
+- subset
 ---
+This dataset contains the original IMDB train split plus balanced dev and test splits derived from the IMDB test set.
+
+### Generation Command
+```bash
+python take_split.py --target_dataset imdb-dev-test-split --dev_size 15000 --test_size 10000 --seed 42 --private
+```
+
+## Dataset Splits
+
+- Train: 25000 samples (label distribution: {0: 12500, 1: 12500})
+- Dev: 15000 samples (label distribution: {0: 7500, 1: 7500})
+- Test: 10000 samples (label distribution: {0: 5000, 1: 5000})
+- Seed: 42
+
+## Notes
+
+- Source: [imdb](https://huggingface.co/datasets/imdb)
+- Train split is the original unmodified IMDB train set (25k samples)
+- Dev and test are carved from the original IMDB test set (25k samples)
+- Dev and test are both label-balanced (50% positive / 50% negative)
+- Each document has a unique MD5 hash ID derived from its text content
+- Dataset passed determinism sanity check before upload
+
+## License
+
+See original dataset (MIT)