Upload params/agent.yaml with huggingface_hub
Browse files- params/agent.yaml +47 -0
params/agent.yaml
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
seed: 42
|
| 2 |
+
num_steps_per_env: 24
|
| 3 |
+
max_iterations: 10001
|
| 4 |
+
obs_groups:
|
| 5 |
+
policy: !!python/tuple
|
| 6 |
+
- policy
|
| 7 |
+
critic: !!python/tuple
|
| 8 |
+
- critic
|
| 9 |
+
save_interval: 100
|
| 10 |
+
experiment_name: go2_velocity
|
| 11 |
+
run_name: ''
|
| 12 |
+
logger: tensorboard
|
| 13 |
+
wandb_project: mjlab
|
| 14 |
+
wandb_tags: !!python/tuple []
|
| 15 |
+
resume: false
|
| 16 |
+
load_run: .*
|
| 17 |
+
load_checkpoint: model_.*.pt
|
| 18 |
+
clip_actions: null
|
| 19 |
+
class_name: OnPolicyRunner
|
| 20 |
+
policy:
|
| 21 |
+
init_noise_std: 1.0
|
| 22 |
+
noise_std_type: scalar
|
| 23 |
+
actor_obs_normalization: true
|
| 24 |
+
critic_obs_normalization: true
|
| 25 |
+
actor_hidden_dims: !!python/tuple
|
| 26 |
+
- 512
|
| 27 |
+
- 256
|
| 28 |
+
- 128
|
| 29 |
+
critic_hidden_dims: !!python/tuple
|
| 30 |
+
- 512
|
| 31 |
+
- 256
|
| 32 |
+
- 128
|
| 33 |
+
activation: elu
|
| 34 |
+
algorithm:
|
| 35 |
+
num_learning_epochs: 5
|
| 36 |
+
num_mini_batches: 4
|
| 37 |
+
learning_rate: 0.001
|
| 38 |
+
schedule: adaptive
|
| 39 |
+
gamma: 0.99
|
| 40 |
+
lam: 0.95
|
| 41 |
+
entropy_coef: 0.01
|
| 42 |
+
desired_kl: 0.01
|
| 43 |
+
max_grad_norm: 1.0
|
| 44 |
+
value_loss_coef: 1.0
|
| 45 |
+
use_clipped_value_loss: true
|
| 46 |
+
clip_param: 0.2
|
| 47 |
+
normalize_advantage_per_mini_batch: false
|