OpenSkillIndex← back to search
unknown

ovmm

@package _global_

install
source · Clone the upstream repo
git clone https://github.com/m-and-m-lab/helios
manifest: configs/ovmm/rl_cont_skill.yaml
source content

@package global

defaults:

habitat_baselines: verbose: False trainer_name: "ddppo" torch_gpu_id: 0 tensorboard_dir: "tb" video_dir: "video_dir" video_fps: 30 test_episode_count: -1 eval_ckpt_path_dir: "data/new_checkpoints"

26 environments will just barely be below 16gb.

20 environments will just barely be below 11gb.

num_environments: 20 checkpoint_folder: "data/new_checkpoints" num_updates: -1 total_num_steps: 2.0e8 log_interval: 10 num_checkpoints: 20

Force PyTorch to be single threaded as

this improves performance considerably

force_torch_single_threaded: True eval_keys_to_include_in_name: ["reward", "force", "success"]

eval: video_option: ["disk"]

rl: policy: name: "PointNavResNetPolicy" action_distribution_type: "gaussian" no_downscaling: True action_dist: use_log_std: True clamp_std: True std_init: -1.0 use_std_param: True max_log_std: 2 obs_transforms: resize_shortest_edge: type: ResizeShortestEdge size: 120 channels_last: True trans_keys: - head_rgb - head_depth - head_panoptic - object_segmentation - receptacle_segmentation - start_recep_segmentation - goal_recep_segmentation - ovmm_nav_goal_segmentation semantic_keys: - head_panoptic - object_segmentation - receptacle_segmentation - start_recep_segmentation - goal_recep_segmentation - ovmm_nav_goal_segmentation center_cropper: type: CenterCropper height: 160 width: 120 channels_last: True trans_keys: - head_rgb - head_depth - head_panoptic - object_segmentation - receptacle_segmentation - start_recep_segmentation - goal_recep_segmentation - ovmm_nav_goal_segmentation

ppo:
  # ppo params
  clip_param: 0.2
  ppo_epoch: 2
  num_mini_batch: 2
  value_loss_coef: 0.5
  entropy_coef: 0.001
  lr: 3e-4
  eps: 1e-5
  max_grad_norm: 0.2
  num_steps: 128
  use_gae: True
  gamma: 0.99
  tau: 0.95
  use_linear_clip_decay: False
  use_linear_lr_decay: False
  reward_window_size: 50

  use_normalized_advantage: False

  hidden_size: 512

  # Use double buffered sampling, typically helps
  # when environment time is similar or larger than
  # policy inference time during rollout generation
  use_double_buffered_sampler: False

ddppo:
  sync_frac: 0.6
  # The PyTorch distributed backend to use
  distrib_backend: NCCL
  # Visual encoder backbone
  pretrained_weights: data/ddppo-models/gibson-2plus-resnet50.pth
  # Initialize with pretrained weights
  pretrained: False
  # Initialize just the visual encoder backbone with pretrained weights
  pretrained_encoder: False
  # Whether the visual encoder backbone will be trained.
  train_encoder: True
  # Whether to reset the critic linear layer
  reset_critic: True

  # Model parameters
  backbone: resnet18
  rnn_type: LSTM
  num_recurrent_layers: 2