AgentGYM proposes the AgentEvol algorithm, enabling the model to evolve itself.
Self-evolution requires the model to explore in the environmens. Please refer to 01-evaluation and README.md in each environment server folder to start the corresponding environment server. If you launches multiple servers, please note the port numbers.
agentenv/examples/agentevol gives a reference implementation of the self-evolution algorithm. Please refer to examples/agentevol/train_agentevol_multi_task.sh to start training.