diff --git a/safepo/multi_agent/macpo.py b/safepo/multi_agent/macpo.py
index 27a034c..60a8fca 100644
--- a/safepo/multi_agent/macpo.py
+++ b/safepo/multi_agent/macpo.py
@@ -254,11 +254,11 @@ def trpo_update(self, sample):
 
         g_step_dir = self.conjugate_gradient(
             self.policy.actor, obs_batch, rnn_states_batch, actions_batch, masks_batch,\
-            available_actions_batch, active_masks_batch, reward_loss_grad.data, nsteps=10
+            available_actions_batch, active_masks_batch, reward_loss_grad.data, nsteps=self.config["conjugate_gradient_iters"]
         )  
         b_step_dir = self.conjugate_gradient(
             self.policy.actor, obs_batch, rnn_states_batch, actions_batch, masks_batch,\
-            available_actions_batch, active_masks_batch, B_cost_loss_grad.data, nsteps=10
+            available_actions_batch, active_masks_batch, B_cost_loss_grad.data, nsteps=self.config["conjugate_gradient_iters"]
         )  
 
         q_coef = (reward_loss_grad * g_step_dir).sum(0, keepdim=True)  
diff --git a/safepo/multi_agent/marl_cfg/macpo/config.yaml b/safepo/multi_agent/marl_cfg/macpo/config.yaml
index cf5ab01..007bf0b 100644
--- a/safepo/multi_agent/marl_cfg/macpo/config.yaml
+++ b/safepo/multi_agent/marl_cfg/macpo/config.yaml
@@ -34,6 +34,7 @@ use_proper_time_limits: False
 
 target_kl: 0.016
 searching_steps: 10
+conjugate_gradient_iters: 10
 accept_ratio: 0.5
 clip_param: 0.2
 learning_iters: 5
@@ -78,5 +79,5 @@ mamujoco:
   gamma: 0.99
   safety_gamma: 0.2
   target_kl: 0.01
-  learning_iters: 15 # Conjugate Gradient Iterations
+  learning_iters: 15 # Number of SGD Iterations
   entropy_coef: 0.01
\ No newline at end of file