NOTE: This is an initial $\beta$-release. We plan to release subsequent versions of Updesh with expanded coverage and enhanced quality control. Future iterations will include larger datasets, improved filtering pipelines.
Updesh is a large-scale synthetic dataset designed to advance post-training of LLMs for Indic languages. It integrates translated reasoning data and synthesized open-domain… See the full description on the dataset page:
https://huggingface.co/datasets/microsoft/Updesh_beta.
