Together Fine-Tuning Platform expands with DPO and continued training support

AI Impact Summary

Together AI has significantly expanded its Fine-Tuning Platform with new capabilities focused on continuous adaptation. The introduction of Direct Preference Optimization (DPO) allows for training models directly on user preference data, moving beyond traditional supervised fine-tuning. Furthermore, the platform now supports resuming training from checkpoints and training on a wider range of open-weight models like Gemma 3 and DeepSeek-R1, alongside improvements to data preprocessing for faster training speeds.