trlx/configs/sweeps/ppo_sweep.yml

tune_config:
  mode: "max"
  metric: "reward/mean"
  search_alg: "random"
  scheduler: "fifo"
  num_samples: 32

# https://docs.ray.io/en/latest/tune/api_docs/search_space.html#tune-sample-docs
optimizer.kwargs.lr:
  strategy: "loguniform"
  values: [0.000001, 0.001]
method.init_kl_coef:
  strategy: "loguniform"
  values: [0.0001, 0.2]
model.num_layers_unfrozen:
  strategy: "choice"
  values: [-1, 2, 6]
method.num_rollouts:
  strategy: "choice"
  values: [32, 128, 512]
method.target:
  strategy: "choice"
  values: [null, 1]

# disable checkpointing for storage sake
train.checkpoint_interval:
  strategy: "choice"
  values: [10000000]
train.save_best:
  strategy: "choice"
  values: [false]