Update README.md

README.md

@@ -1,21 +1,145 @@
 ---
 dataset_info:
   features:
-  - name: id
-    dtype: string
-  - name: input
-    dtype: string
-  - name: output
-    dtype: string
+    - name: id
+      dtype: string
+    - name: input
+      dtype: string
+    - name: output
+      dtype: string
   splits:
-  - name: train
-    num_bytes: 37238584810
-    num_examples: 12500000
-  download_size: 71279710676
-  dataset_size: 37238584810
+    - name: train
+      num_bytes: 29197203208
+      num_examples: 12500000
+  download_size: 11862158116
+  dataset_size: 29197203208
 configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*
+license: cc-by-4.0
+language:
+- en
+tags:
+- code
+- synthetic
+- instruction-tuning
+task_categories:
+- text-generation
+size_categories:
+- 10M<n<100M
 ---
+
+# ultimate-code
+
+## Dataset Description
+
+ultimate-code is a derived dataset built by combining and processing data from
+[`nvidia/OpenCodeInstruct`](https://huggingface.co/datasets/nvidia/OpenCodeInstruct) and
+[`nvidia/OpenCodeGeneticInstruct`](https://huggingface.co/datasets/nvidia/OpenCodeGeneticInstruct).
+It can be used to fine-tune LLMs for coding tasks.
+
+> **Tokenized variant available:** a pre-tokenized version of this dataset is available at
+> [`CodeForCodersYT/ultimate-code-tokenized`](https://huggingface.co/datasets/CodeForCodersYT/ultimate-code-tokenized).
+> <!-- TODO: Link/Repo-Namen prüfen und ggf. anpassen, falls der Name der tokenisierten Variante anders lautet. -->
+> Use that version if you want ready-to-train tokenized sequences instead of raw text.
+
+## Source Datasets & Attribution
+
+This dataset is derived from the following NVIDIA datasets, both released under the
+**Creative Commons Attribution 4.0 International License (CC BY 4.0)**:
+
+| Source Dataset | Owner | License | Link |
+|---|---|---|---|
+| OpenCodeInstruct | NVIDIA Corporation | CC BY 4.0 | https://huggingface.co/datasets/nvidia/OpenCodeInstruct |
+| OpenCodeGeneticInstruct | NVIDIA Corporation | CC BY 4.0 | https://huggingface.co/datasets/nvidia/OpenCodeGeneticInstruct |
+
+Both source datasets are copyright © NVIDIA Corporation and are licensed under CC BY 4.0
+([full license text](https://creativecommons.org/licenses/by/4.0/legalcode)).
+
+### Modifications made to the original data
+
+- Both source datasets were merged into a single unified schema.
+- Only the `id`, `input`, and `output` columns were kept; all other columns from the original
+  datasets (e.g. `domain`, `generation_algorithm`, `llm_judgement`, `unit_tests`,
+  `tests_execution_status`, `average_test_score`) were dropped.
+- No additional filtering, deduplication, or reformatting was applied beyond the column selection.
+
+## Dataset Structure
+
+| Field | Type | Description |
+|---|---|---|
+| `id` | string | Unique identifier for each example |
+| `input` | string | The coding question / prompt |
+| `output` | string | The corresponding solution / response |
+
+## Dataset Statistics
+
+- Total records (train split): 12,500,000
+- Total size (uncompressed): ~29.2 GB
+- Download size (Parquet): ~11.86 GB
+
+## How to Use
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("CodeForCodersYT/ultimate-code", split="train")
+```
+
+For a pre-tokenized version ready for training pipelines, see
+[`CodeForCodersYT/ultimate-code-tokenized`](https://huggingface.co/datasets/CodeForCodersYT/ultimate-code-tokenized) instead.
+
+## License
+
+This dataset is released under **CC BY 4.0** (https://creativecommons.org/licenses/by/4.0/legalcode),
+consistent with the license of the source datasets. If you use this dataset, please provide
+attribution to both this dataset and the original NVIDIA source datasets as described above.
+
+## Ethical Considerations
+
+This dataset inherits ethical considerations from its source data. As noted by NVIDIA for the
+original datasets: users are responsible for checking that the dataset and license are fit for
+their intended purpose, and should evaluate the resulting models for their specific use case and
+industry requirements before deployment.
+
+## Citation
+
+If you use this dataset, please cite both the original source datasets and this dataset.
+
+**OpenCodeInstruct:**
+
+```bibtex
+@article{ahmad2025opencodeinstruct,
+      title={OpenCodeInstruct: A Large-scale Instruction Tuning Dataset for Code LLMs},
+      author={Wasi Uddin Ahmad and Aleksander Ficek and Mehrzad Samadi and Jocelyn Huang and Vahid Noroozi and Somshubra Majumdar and Boris Ginsburg},
+      year={2025},
+      eprint={2504.04030},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/2504.04030},
+}
+```
+
+**OpenCodeGeneticInstruct:**
+
+```bibtex
+@misc{nvidia_opencodegeneticinstruct,
+      title={OpenCodeGeneticInstruct},
+      author={NVIDIA Corporation},
+      year={2025},
+      url={https://huggingface.co/datasets/nvidia/OpenCodeGeneticInstruct},
+}
+```
+
+**This dataset:**
+
+```bibtex
+@misc{ultimate_code_2026,
+      title={ultimate-code: A Merged Instruction-Tuning Dataset for Code LLMs},
+      author={CodeForCodersYT},
+      year={2026},
+      url={https://huggingface.co/datasets/CodeForCodersYT/ultimate-code},
+}
+```