4. The pre-trained model can act as a very good starting point letting good-tuning to converge more rapidly than training from scratch. 1. We introduce AntEval, a novel framework tailored for that evaluation of interaction capabilities in LLM-driven agents. This framework introduces an intera