malr07/opc-sft-stage2-dense-extracted

--- dataset_info: - config_name: educational_instruct features: - name: seq_id dtype: int64 - name: instruction dtype: string - name: output dtype: string - name: code dtype: string - name: entry_point dtype: string - name: testcase sequence: string - name: dense_snippets sequence: string - name: snippets_length_ratio dtype: float64 splits: - name: train num_bytes: 145475621 num_examples: 118278 download_size: 62934675 dataset_size: 145475621 - config_name: evol_instruct features: - name: instruction dtype: string - name: output dtype: string - name: tag dtype: string - name: dense_snippets sequence: string - name: snippets_length_ratio dtype: float64 splits: - name: train num_bytes: 293103162 num_examples: 111183 download_size: 163590666 dataset_size: 293103162 - config_name: mceval_instruct features: - name: instruction dtype: string - name: output dtype: string - name: tag dtype: string - name: dense_snippets sequence: string - name: snippets_length_ratio dtype: float64 splits: - name: train num_bytes: 154776402 num_examples: 35943 download_size: 73019183 dataset_size: 154776402 - config_name: package_instruct features: - name: instruction dtype: string - name: output dtype: string - name: tag dtype: string - name: dense_snippets sequence: string - name: snippets_length_ratio dtype: float64 splits: - name: train num_bytes: 658293249 num_examples: 170943 download_size: 315933015 dataset_size: 658293249 configs: - config_name: educational_instruct data_files: - split: train path: educational_instruct/train-* - config_name: evol_instruct data_files: - split: train path: evol_instruct/train-* - config_name: mceval_instruct data_files: - split: train path: mceval_instruct/train-* - config_name: package_instruct data_files: - split: train path: package_instruct/train-* license: mit task_categories: - text-generation - question-answering size_categories: - 100K<n<1M ---  # OpenCoder Dataset Dense Region Extracted This dataset is a post-processed version of the OpenCoder SFT Stage2 dataset ([opc-sft-stage2](https://huggingface.co/datasets/OpenCoder-LLM/opc-sft-stage2)). We use gpt-4o API to extract the information dense regions from each sample and logged them in the `dense_snippets` column.Detailed information about the data can be found in our [paper](https://arxiv.org/abs/2603.15803). ## OpenCoder's sft-stage2 summary The original version of this dataset is used in OpenCoder's Stage 2 and consists of four parts: * **educational_instruct**: Using the [algorithmic corpus](https://huggingface.co/datasets/OpenCoder-LLM/opc-annealing-corpus) as a seed, OpenCoder generated (instruction, code, test case) triples, validated through a Python compiler. Notably, the inclusion of test cases provides a valuable signal for code RL. * **evol_instruct**: Directly using the open-source version [MagicCoder-Evol-Instruct-110k](https://huggingface.co/datasets/ise-uiuc/Magicoder-Evol-Instruct-110K). * **mceval_instruct**: Directly using the open-source version [McEval-Instruct](https://huggingface.co/datasets/Multilingual-Multimodal-NLP/McEval-Instruct). * **package_instruct**: OpenCoder extracted common interface documentation from pydoc and used it as a seed to generate Python package-related questions. ## How to use it ```python from datasets import load_dataset repo_id = "malr07/opc-sft-stage2-dense-extracted" educational_instruct = load_dataset(repo_id, "educational_instruct") evol_instruct = load_dataset(repo_id, "evol_instruct") mceval_instruct = load_dataset(repo_id, "mceval_instruct") package_instruct = load_dataset(repo_id, "package_instruct") ``` ## Citation Information Please consider citing our [paper](https://arxiv.org/abs/2603.15803) if you find this dataset useful: ``` @misc{ma2026maskdllmneedsmasked, title={Mask Is What DLLM Needs: A Masked Data Training Paradigm for Diffusion LLMs}, author={Linrui Ma and Yufei Cui and Kai Han and Yunhe Wang}, year={2026}, eprint={2603.15803}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2603.15803}, } ```