Update README.md

README.md

@@ -26,6 +26,71 @@ configs:
     path: data/test-*
 ---
 
+# Dataset Card for Small C4 Dataset (10k Train, 10k Validation, 10k Test)
+
+## Dataset Summary
+
+The **Small C4 Dataset** is a reduced version of the original [C4 dataset](https://huggingface.co/datasets/allenai/c4) (Colossal Clean Crawled Corpus), designed to facilitate lightweight experimentation and model training without the need to process the full C4 dataset. This dataset includes:
+- **10,000 examples** for training,
+- **10,000 examples** for validation, and
+- **10,000 examples** for testing.
+
+Each example consists of a single text passage, sourced from the English subset of the original C4 corpus.
+
+## Dataset Details
+
+- **Source**: [allenai/c4](https://huggingface.co/datasets/allenai/c4)
+- **Subset Language**: English
+- **Streaming Enabled**: Yes (streaming=True used to sample without downloading the entire dataset)
+- **Sampling Method**:
+  - **Training Set**: First 10,000 examples from the `train` split of C4.
+  - **Validation Set**: First 10,000 examples from the `validation` split of C4.
+  - **Test Set**: The next 10,000 examples from the `validation` split (after the validation set).
+- **Dataset Size**: 30,000 examples in total.
+
+## Dataset Creation
+
+The dataset was created using Hugging Face’s `datasets` library with streaming enabled to handle the large size of the original C4 dataset efficiently. A subset of examples was sampled in parallel for each of the train, validation, and test splits.
+
+## Usage
+
+This dataset is suitable for lightweight model training, testing, and experimentation, particularly useful when:
+- **Computational resources** are limited,
+- **Prototyping** models before scaling to the full C4 dataset, or
+- **Evaluating** model performance on a smaller, representative sample of the full corpus.
+
+## Example Usage
+
+```python
+from datasets import load_dataset
+
+# Load the small C4 dataset
+dataset = load_dataset("brando/small-c4-dataset")
+
+# Access train, validation, and test splits
+train_data = dataset["train"]
+validation_data = dataset["validation"]
+test_data = dataset["test"]
+
+# Example: Display a random training example
+print(train_data[0])
+License
+This dataset inherits the licensing of the original C4 dataset, which follows the Apache License 2.0.
+
+Citation
+If you use this dataset in your work, please cite the original C4 dataset or my ultimate utils repo:
+
+```
+@misc{miranda2021ultimateutils,
+    title={Ultimate Utils - the Ultimate Utils Library for Machine Learning and Artificial Intelligence},
+    author={Brando Miranda},
+    year={2021},
+    url={https://github.com/brando90/ultimate-utils},
+    note={Available at: \url{https://www.ideals.illinois.edu/handle/2142/112797}},
+    abstract={Ultimate Utils is a comprehensive library providing utility functions and tools to facilitate efficient machine learning and AI research, including efficient tensor manipulations and gradient handling with methods such as `detach()` for creating gradient-free tensors.}
+}
+```
+
 Script that created it
 
 ```python
@@ -93,4 +158,4 @@ def main() -> None:
 if __name__ == "__main__":
     main()
 
-```
+```