Danfoa / commons_game

A custom implementation of DeepMind's "the commons game"
9 stars 0 forks source link

Question Relate to Platform Set Up and Recommendation for This Code Implementation #2

Closed XinqiGao closed 2 months ago

XinqiGao commented 5 months ago

Hi, Daniel!

I'm interested in this commons game and your implementation. I have my MacBook Pro with M2 chip as my laptop and wondering if that's enough to run the complete game locally? Do you recommend using CPU or GPU to run this experiment with multiple agent? Or it's better to run the game in clouds like Google Colab?

The problem I'm current facing is that: I successfully set up the TensorFlow with plugins which is optimized for the Apple Silicon M1/M2 chips. I also verified and tested that the GPU is available and my tensor flow is indeed utilizing the GPU.  But the code is ran really slow with generating images and videos. It takes around 30 minutes to generate the first episode 700 images and I have the 137 exit code which saying the memory exceeds the memory limit, and it stopped the program running.

Also, I see there is a Google Colab link but it's no longer available and I cannot open it at all. Could you reactivate the file if possible so that I can check the implementation and see how large the experiment could be?

Best Regards, Xinqi