Update README.md

README.md

@@ -1,3 +1,45 @@
 ---
 license: apache-2.0
+size_categories:
+- 10M<n<100M
 ---
+
+# Dataset Card: Sudoku Dataset
+
+## Dataset Overview
+
+- **Dataset Sources:**
+
+  1. [1 million Sudoku games](https://www.kaggle.com/datasets/bryanpark/sudoku)
+  2. [3 million Sudoku puzzles with ratings](https://www.kaggle.com/datasets/radcliffe/3-million-sudoku-puzzles-with-ratings)
+  3. [4 Million Sudoku Puzzles Easy-to-Hard](https://www.kaggle.com/datasets/informoney/4-million-sudoku-puzzles-easytohard)
+  4. [9 Million Sudoku Puzzles and Solutions](https://www.kaggle.com/datasets/rohanrao/sudoku)
+  5. ~2k miscellaneous scraped puzzles
+
+- **Dataset Size:** 17M puzzles (16.7M for training, 300K for evaluation)
+
+## Data Format
+
+- **File Format:** Parquet files
+- **Dataset Split:**
+  - `train_data`: 2.7M puzzles
+  - `valid_data`: 0.3M puzzles
+  - `train_aux`: 14M puzzles
+  - `valid_aux`: 2k puzzles
+
+## Dataset Attributes
+
+The dataset contains the following attributes:
+
+- `puzzle`: Represents the Sudoku puzzle configuration.
+- `solution`: Corresponding solutions to the puzzles.
+- `difficulty`: Indicates the difficulty level of the puzzles.
+- `source`: Source identifier for the puzzle (e.g., dataset name or origin).
+
+## Data Preprocessing
+
+- **Handling Blanks:** Blanks in the puzzles are represented as '0'.
+
+## Licensing and Legal Considerations
+
+- The dataset is sourced from various Kaggle datasets, and each source may have its own licensing terms. Please refer to the respective Kaggle pages for licensing information.