Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Models | Read Paper on Bytez