ML-Agents 4.0.0 is here! This release brings support for the powerful Inference Engine and sets the stage for the future of ML-Agents in Unity. Developer documentation has now moved to the Unity Package documentation, your new go-to resource for everything ML-Agents.
Awesome - will test this out today!
EDIT: Very cool, works well so far:
6 Likes
Hey @maryam_zia thanks for the update, happy to see ML Agents progressing 
I’ve tried latest version on two different machines:
- Windows + Nvidia machine, has issue with mixing tensors on on cuda and gpu if threaded is true in config. Could be an easy fix, sadly don’t have much time atm to look into it deeper.
- On mac machine I hit some performance regression issue. Training slowed down 6 times compared to my previous version (Unity 2022.3 + ML-Agents 3.0.0-exp.1). My guess is that for some reason training happens on CPU instead of GPU. Could be issues with my setup. Will update the thread if I find a solution.
Created PR for Threaded: true (cuda-cpu mixing) issue
2 Likes
Are there examples for Unity 6000.3?? The examples don’t work, only 3d ball compiles.
I can confirm that Mac performance regression seems real.
I’ve tested on custom toy environment and got 5-7x difference in profile for DecideAction call.
I intentionally created quite big policy and value networks to highlight the difference on 512 simultaneous training areas.
Unity 2022.3.10f1
com.unity.ml-agents 3.0.0-exp.1
com.unity.ml-agents.extensions 0.6.1-exp.1
ml-agents: 1.0.0,
ml-agents-envs: 1.0.0,
Communicator API: 1.5.0,
PyTorch: 1.13.1
Unity 6000.2.4f1
com.unity.ml-agents 4.0.0
ml-agents: 1.2.0.dev0,
ml-agents-envs: 1.2.0.dev0,
Communicator API: 1.5.0,
PyTorch: 2.2.2
If anybody want to test it, here is the repo with both project GitHub - bmind7/MLAgents4MacIssue
To Reproduce
Steps to reproduce the behavior:
- Go to project
./MLAgents/folder in terminal - Start
mlagents-learn config.yaml --run-id="platform_test" --time-scale=1 --force - Start scene Dojo in unity
- Profile
1 Like
Hi, please check out our examples in the repo: ml-agents/Project/Assets/ML-Agents/Examples at develop · Unity-Technologies/ml-agents · GitHub
Hi, thanks for reporting this. What are the specs of your machine? Does this regression happen only on mac or windows as well?
Hi, sorry forgot about machine
Here it is
I don’t see any performance regression on Windows + Nvidia GPU machine
Another little sample with 4.0.0:
2 Likes
wonderful!!! keep going!
is it dots compatible ?