2025
Skrull: Towards Efficient Long Context Fine-tuning through Dynamic Data Scheduling
NeurIPS 2025poster
Long-context supervised fine-tuning (Long-SFT) plays a vital role in enhancing the performance of large language models (LLMs) on long-context tasks. To smoothly adapt LLMs to long-context scenarios, this process typically entails training on mixed datasets containing both long and short sequences.…