Search is not available for this dataset
norm_stats dict |
|---|
{
"state": {
"mean": [
-0.5322057604789734,
0.29820895195007324,
-0.3608394265174866,
0.46594980359077454,
1.109459638595581,
0.10193049162626266,
-0.30093124508857727,
0.21558423340320587,
-0.14903324842453003,
-1.181117296218872,
4.38492631912231... |
{
"state": {
"mean": [
-0.5322057604789734,
0.29820895195007324,
-0.3608394265174866,
0.46594980359077454,
1.109459638595581,
0.10193049162626266,
-0.30093124508857727,
0.21558423340320587,
-0.14903324842453003,
-1.181117296218872,
4.38492631912231... |
{
"state": {
"mean": [
-0.5322057604789734,
0.29820895195007324,
-0.3608394265174866,
0.46594980359077454,
1.109459638595581,
0.10193049162626266,
-0.30093124508857727,
0.21558423340320587,
-0.14903324842453003,
-1.181117296218872,
4.38492631912231... |
G1_23 restock Pi0.5 trimmed checkpoints
Inference-ready Orbax checkpoints from the independent restock_trim_v2 experiment. The leading idle period was trimmed per episode; the original train/validation/test episode assignment was retained.
Directories: trim_v2/12000, trim_v2/14000, trim_v2/16000. Each contains params/, assets/ (normalization statistics), and checkpoint metadata. Optimizer train_state/ is not included. The validation-selected checkpoint is trim_v2/12000. See manifest.json for validation metrics.
Load with the pi05_g1_23_restock_trim_v2 configuration in the corresponding openpi-fintune-trim-v2 code, pointing create_trained_policy at the downloaded checkpoint directory. These offline metrics do not measure robot task success.
- Downloads last month
- 141