python train.py --local-mode true
python train.py --run-name <YOUR_RUN_NAME> --device <YOUR_DEVICE> --seed <YOUR_NUMBER> --num-envs 1 --rollout-batch-size 2**14
python evaluate.py -p policies
python evaluate.py -p <YOUR_DIR> -r
NMMO Baseline Repository:
├── reinforcement_learning
│ ├── config.py
│ └── policy.py --> Your policy goes here
├── requirements.txt
└── train.py --> Train your policy here
- inference with apache beam
- https://engineering.linkedin.com/teams/data/data-infrastructure/machine-learning-infrastructure
- https://www.linkedin.com/pulse/machine-learning-platforms-ai-lifecycle-management-mark-mondoka-holtc/
- FastTransformers
- Deepspeed
- https://www.tensorflow.org/tfx
- Ray, KubeFlow or W&B