Published on September 22, 2026, iSDFT describes a continual-learning method for LLMs based on on-policy self-distillation and information proximity. Its repository provides training code, checkpoints, and evaluation results.
Published on September 22, 2026, iSDFT presents a continual-learning method for large language models, described as on-policy self-distillation with information proximity. The material concerns an approach for exploring ongoing updates to these models.
The project repository provides training code, checkpoints, and evaluation results. To consult and verify the claims, inspect these materials and check whether the results match the configuration and metrics reported in the repository. If you use AI to study or apply the method, avoid sending personal data or confidential content without authorization, and review your organization's data-handling policies.