RL agent using private and shared world models
Requirements:
- Linux/Windows: CUDA Drivers installed (compatibility matrix)
- Linux: Docker Engine, nvidia-docker2 installed, "nvidia-smi" docker test on previous link needs to work.
- Windows:
- WSL nbody benchmark needs to work
- Docker Desktop must be installed even if you intend on using docker from the command prompt in WSL.
- "Does nvidia-docker2 support Windows?"
$ git clone https://github.com/neureal/neureal-ai-agent.git
$ cd neureal-ai-agent
$ ./build.sh build
$ ./build.sh dev
root@f3537fad7977:/app# time python agent.py
...
RUN gym-PG-CartPole-r-dev-a0-22-10-03-11-10-04
tracing -> GeneralAI PG
tracing -> GeneralAI pg_actor
tracing -> GeneralAI pg_learner_onestep
RUNNING
gym-PG-CartPole-r-dev-a0-22-10-03-11-10-04 [CPU-float64] A[inAio+D512-Ï7_net02AT+D256_outAio+D512-Öc2_Ld7x16]-new
time: 0:00:00:21 steps:2712 t/s:0.00781960 ms:256 | attn: net io out ar al:16 am:4 | a-clk:0.001 a-spd:160.0 aug:SP | action:4e-06
real 0m28.174s
user 0m42.520s
sys 0m9.661s
root@f3537fad7977:/app#