Equidiff / equidiff /equi_diffpo /config /train_act_abs.yaml
Lillianwei's picture
mimicgen
c1f1d32
defaults:
- _self_
- task: mimicgen_abs
name: act
_target_: equi_diffpo.workspace.train_act_workspace.TrainActWorkspace
shape_meta: ${task.shape_meta}
exp_name: "default"
task_name: stack_d1
n_demo: 200
horizon: 10
n_obs_steps: 1
n_action_steps: 10
n_latency_steps: 0
dataset_obs_steps: ${n_obs_steps}
past_action_visible: False
dataset: equi_diffpo.dataset.robomimic_replay_image_dataset.RobomimicReplayImageDataset
dataset_path: data/robomimic/datasets/${task_name}/${task_name}_abs.hdf5
policy:
_target_: equi_diffpo.policy.act_policy.ACTPolicyWrapper
shape_meta: ${shape_meta}
max_timesteps: ${task.env_runner.max_steps}
temporal_agg: false
n_envs: ${task.env_runner.n_envs}
horizon: ${horizon}
dataloader:
batch_size: 64
num_workers: 4
shuffle: True
pin_memory: True
persistent_workers: True
val_dataloader:
batch_size: 64
num_workers: 4
shuffle: False
pin_memory: True
persistent_workers: True
training:
device: "cuda:0"
seed: 0
debug: False
resume: True
num_epochs: ${eval:'50000 / ${n_demo}'}
rollout_every: ${eval:'1000 / ${n_demo}'}
checkpoint_every: ${eval:'1000 / ${n_demo}'}
val_every: 1
max_train_steps: null
max_val_steps: null
tqdm_interval_sec: 1.0
logging:
project: diffusion_policy_${task_name}
resume: True
mode: online
name: act_demo${n_demo}
tags: ["${name}", "${task_name}", "${exp_name}"]
id: null
group: null
checkpoint:
topk:
monitor_key: test_mean_score
mode: max
k: 5
format_str: 'epoch={epoch:04d}-test_mean_score={test_mean_score:.3f}.ckpt'
save_last_ckpt: True
save_last_snapshot: False
multi_run:
run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
hydra:
job:
override_dirname: ${name}
run:
dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
sweep:
dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
subdir: ${hydra.job.num}