SAMA: Towards Multi-Turn Referential Grounded Video Chat with Large Language Models | Read Paper on Bytez