The current version of the code already allows to run the game with Docker, but there are some issues with loading the agents. To solve this, there are 2 options:
upgrade the code to use tf 2.6.0 and ray 1.1.0. To do this, the main problem comes for the class of Recurrent Network (it has different names, location, as well as some updates). I tried to change the names, but it seems to require further debugging.
another option is to rebuild ray 0.8.5 with the support of m1 architecture. This can be done with bazel. I have added part of the commands for installing bazel when compiling the docker image, additionally the code for the package rebuild part should be added. If bazel crashes during the build, try using the -j 1(2,4,8) options as it tends to run out of memory
The current version of the code already allows to run the game with Docker, but there are some issues with loading the agents. To solve this, there are 2 options: