File size: 2,107 Bytes
94f697a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 | defaults:
- env_vars: env_vars
- encoder: dino_patch
- _self_
- cluster: local
hydra:
run:
dir: ${env_vars.save_path}/outputs/${env.gym.id}/${now:%Y-%m-%d}/${now:%H-%M-%S}
job:
chdir: True
sweep:
dir: ${env_vars.save_path}/outputs/${env.gym.id}/${now:%Y-%m-%d}/${now:%H-%M-%S}
subdir: ${hydra.job.num}
seed: 42
window_size: 2
goal_window_size: 2
eval_window_size: 2
num_workers: 0
batch_size: 32
epochs: 300
eval_freq: 10
eval_on_env_freq: 10
num_env_evals: 100
num_envs: 10
num_final_evals: 100
action_window_size: 3
pred_horizon: ${eval:'${action_window_size} + ${window_size} - 1'}
n_action_steps: ${action_window_size} # Action steps to roll out
precompute_embeddings: False
goal_dim: 0
wandb:
project: patch_policy
entity: ${env_vars.wandb_entity}
train_fraction: 0.9
device: cuda
optim:
lr: 1e-4
weight_decay: 0.0
betas: [0.9, 0.999]
env:
gym:
_target_: envs.cube.cube_env.CubeWrapper
task_id: 5
id: 'cube'
env:
_target_: gymnasium.make
id: 'cube-double-v1'
reward_task_id: 5
max_episode_steps: 300
mode: 'task'
act_dim: 5
goal_dim: ${goal_dim}
views: 1
dataset:
_target_: datasets.cube.CubeDataset
data_directory: ${env_vars.datasets.cube}
data:
window_size: ${window_size}
action_window_size: ${action_window_size}
vqbet_get_future_action_chunk: false
goal_conditional: future
future_seq_len: ${goal_window_size}
min_future_sep: ${action_window_size}
save_ckpt: true
save_path: "${env_vars.save_path}/outputs/${env.gym.id}/${now:%Y-%m-%d}/${now:%H-%M-%S}"
load_path: null
model:
_target_: models.diffusion_policy.diffusion_policy.DiffusionPolicy
obs_dim: ${encoder.output_dim}
act_dim: ${env.act_dim}
goal_dim: ${env.goal_dim}
views: ${env.views}
obs_horizon: ${window_size}
pred_horizon: ${pred_horizon}
action_horizon: ${n_action_steps}
visual_input: true
data_act_scale: 1
lr: ${optim.lr}
weight_decay: ${optim.weight_decay}
use_transform: null
n_patches: ${encoder.n_patches} # number of patches along one dimension (e.g. 16 for 16x16 patches)
|