# How to Implement Decentralized Execution

**URL:** <https://discuss.ray.io/t/how-to-implement-decentralized-execution/257>\
**Category:** RLlib\
**Created:** [December 16, 2020, 3:50pm UTC](https://discuss.ray.io/t/how-to-implement-decentralized-execution/257 "2020-12-16T15:50:54Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![helloSeen](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/helloseen/32/163_2.png) [@helloSeen](https://discuss.ray.io/u/helloSeen)\
**Post date:** [December 16, 2020, 3:50pm UTC](https://discuss.ray.io/t/how-to-implement-decentralized-execution/257/1 "2020-12-16T15:50:54Z")

</div>

Hi all,

This may be a very simple question, but I am having a hard time understanding this. I am following along with the Two-Step Game environment for multi-agent reinforcement learning. I am training with a Centralized Critic with PPO (see: [https://github.com/ray-project/ray/blob/master/rllib/examples/centralized\_critic.py](https://github.com/ray-project/ray/blob/master/rllib/examples/centralized_critic.py))

After I train the model, how do I deploy it for execution only? In the same vein, I have also looked at using QMIX (see: [https://github.com/ray-project/ray/blob/master/rllib/examples/two\_step\_game.py](https://github.com/ray-project/ray/blob/master/rllib/examples/two_step_game.py)) where the agents are grouped during training. Should I ungroup them during execution?

Thanks for the help!

---

<div class="post-metadata">

**Author:** ![hermmanhender](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/hermmanhender/32/5691_2.png) [@hermmanhender](https://discuss.ray.io/u/hermmanhender)\
**Post date:** [September 6, 2023, 2:31am UTC](https://discuss.ray.io/t/how-to-implement-decentralized-execution/257/2 "2023-09-06T02:31:54Z")

</div>

Hi! I’m working with Qmix and I have the same question.  
Is it possible to make a decentralized explotation of a trained Qmix algorithm?

I don’t find any information about it in the docs or discuss forum. Any ideas are welcome to understand better Qmix algorithm 🤗

---

<div class="post-metadata">

**Author:** ![nkotecha](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/nkotecha/32/5131_2.png) [@nkotecha](https://discuss.ray.io/u/nkotecha)\
**Post date:** [September 28, 2023, 12:26pm UTC](https://discuss.ray.io/t/how-to-implement-decentralized-execution/257/3 "2023-09-28T12:26:36Z")

</div>

Hey, I’m also trying to implement decentralised execution after training my MAPPO policy. Have you managed to figure out how to do it? Thanks

---

<div class="post-metadata">

**Author:** ![SExpert12](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/sexpert12/32/6311_2.png) [@SExpert12](https://discuss.ray.io/u/SExpert12)\
**Post date:** [February 18, 2025, 7:17am UTC](https://discuss.ray.io/t/how-to-implement-decentralized-execution/257/4 "2025-02-18T07:17:55Z")

</div>

Does anyone have done this? I am also working on the same.