Kimi-K2-Instruct-open-perfectblend-regenerate

## Dataset Source - **Base Dataset:** Open-PerfectBlend (1.4M samples). - **Generator Model:** Kimi-K2-Instruct. - **Method:** Automated inference aimed at creating a "perfect blend" of instruction following capabilities tailored to the Kimi-K2 architecture. --- ## Dataset Usage This dataset is the specific training corpus for **Kimi-K2-Instruct-eagle3**. It significantly enhances the model's ability to handle multi-task scenarios—including complex mathematical reasoning, code generation, and chat dialogue—by providing high-consistency training signals derived directly from the teacher model. --- ## Citation If you utilize this dataset in your research or application, please cite the following: ```bibtex @misc{kimik22025regenerate, title={Kimi-K2-Instruct-open-perfectblend-regenerate: Synthetic Data for EAGLE3 Acceleration}, author={Ant AQ Team}, year={2025}, }