Will the training code be open-sourced?

[crazy-wq]

Question

Thank you for releasing this great model / codebase. I would like to ask whether the team has plans to open-source the full training code (including data preprocessing, training scripts, configuration files, etc.)?

Motivation

• Open-sourcing the training code helps the community reproduce results, understand technical details, and build further improvements and extensions.
• For academic research and non-commercial applications, accessible training pipelines greatly lower the barrier for further development.
• If full open-sourcing is not possible for some reason, would it be feasible to share core training pipelines or some key scripts?

Thank you for your consideration and contribution!