✨
AI Summary
- Introduces Dynamic Fine-Tuning (DFT) to improve LLM generalization.
- Addresses reward structure issues in standard Supervised Fine-Tuning.
- Dynamically re-weights the objective function for better token probability.