dppo/cfg/furniture/pretrain/one_leg_low/pre_gaussian_mlp.yaml
Allen Z. Ren 1d04211666 v0.7 (#26)
* update from scratch configs

* update gym pretraining configs - use fewer epochs

* update robomimic pretraining configs - use fewer epochs

* allow trajectory plotting in eval agent

* add simple vit unet

* update avoid pretraining configs - use fewer epochs

* update furniture pretraining configs - use same amount of epochs as before

* add robomimic diffusion unet pretraining configs

* update robomimic finetuning configs - higher lr

* add vit unet checkpoint urls

* update pretraining and finetuning instructions as configs are updated
2024-11-20 15:56:23 -05:00

61 lines
1.5 KiB
YAML

defaults:
- _self_
hydra:
run:
dir: ${logdir}
_target_: agent.pretrain.train_gaussian_agent.TrainGaussianAgent
name: ${env}_pre_gaussian_mlp_ta${horizon_steps}
logdir: ${oc.env:DPPO_LOG_DIR}/furniture-pretrain/${name}/${now:%Y-%m-%d}_${now:%H-%M-%S}_${seed}
train_dataset_path: ${oc.env:DPPO_DATA_DIR}/furniture/${task}_${randomness}/train.npz
seed: 42
device: cuda:0
task: one_leg
randomness: low
env: ${task}_${randomness}_dim
obs_dim: 58
action_dim: 10
horizon_steps: 8
cond_steps: 1
wandb:
entity: ${oc.env:DPPO_WANDB_ENTITY}
project: furniture-${task}-${randomness}-pretrain
run: ${now:%H-%M-%S}_${name}
train:
n_epochs: 3000
batch_size: 256
learning_rate: 1e-4
weight_decay: 1e-6
lr_scheduler:
first_cycle_steps: 3000
warmup_steps: 100
min_lr: 1e-5
save_model_freq: 500
model:
_target_: model.common.gaussian.GaussianModel
network:
_target_: model.common.mlp_gaussian.Gaussian_MLP
mlp_dims: [1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]
use_layernorm: False # worse with layernorm
activation_type: ReLU # worse with Mish
residual_style: True
fixed_std: 0.1
cond_dim: ${eval:'${obs_dim} * ${cond_steps}'}
horizon_steps: ${horizon_steps}
action_dim: ${action_dim}
horizon_steps: ${horizon_steps}
device: ${device}
ema:
decay: 0.995
train_dataset:
_target_: agent.dataset.sequence.StitchedSequenceDataset
dataset_path: ${train_dataset_path}
horizon_steps: ${horizon_steps}
cond_steps: ${cond_steps}
device: ${device}