Skip to main content
claire-labo
Projects
no-representation-no-trust-release-cleanrl
Reports
Replication of MuJoCo with CleanRL
Log in
Sign up
Share
Comment
Star
Share
Comment
Star
Replication of MuJoCo with CleanRL
Replicating our TorchRL collapsed MuJoCo runs with CleanRL scripts.
Skander Moalla
Created on October 29
|
Last edited on November 16
Comment
batch/start/SVD/approximate_rank_pca/features_value_batch
batch/start/SVD/approximate_rank_pca/features_value_batch
1M
2M
3M
4M
global_step
5
10
15
20
25
30
batch/start/SVD/approximate_rank_pca/features_policy_batch
batch/start/SVD/approximate_rank_pca/features_policy_batch
1M
2M
3M
4M
global_step
10
20
30
update_epochs: 10
CleanRL
capacity.num_epochs: -
TorchRL
batch/perf/avg_return_raw
batch/perf/avg_return_raw
1M
2M
3M
4M
global_step
0
1000
2000
3000
update_epochs: 10
CleanRL
capacity.num_epochs: -
TorchRL
batch/last_epoch/last_minibatch/loss/loss_policy
batch/last_epoch/last_minibatch/loss/loss_policy
1M
2M
3M
4M
global_step
0
1e+8
2e+8
3e+8
4e+8
batch/last_epoch/last_minibatch/loss/entropy
batch/last_epoch/last_minibatch/loss/entropy
1M
2M
3M
4M
global_step
0
5e+10
1e+11
1.5e+11
TorchRL
5
CleanRL
5
Add a comment