One confusion about DQN

def valid_agent(env_class, env_args: dict, net_dims: List[int], agent_class, actor_path: str, render_times: int = 8):
env = build_env(env_class, env_args)

state_dim = env_args['state_dim']
action_dim = env_args['action_dim']
agent = agent_class(net_dims, state_dim, action_dim, gpu_id=-1)
actor = agent.act

print(f"| render and load actor from: {actor_path}")
actor.load_state_dict(th.load(actor_path, map_location=lambda storage, loc: storage))
for i in range(render_times):
cumulative_reward, episode_step = get_rewards_and_steps(env, actor, if_render=True)
print(f"|{i:4} cumulative_reward {cumulative_reward:9.3f} episode_step {episode_step:5.0f}")

Does the above code read the trained pth(dict) files, namely neural network parameters, and directly test the new environment with the neural network (no longer use the explore rate)? Why do you choose th.save(actor.state_dict(), save_path) instead of direct actor for saving pth files?