Document step_5000 + step_7500 checkpoints

README.md

@@ -78,14 +78,15 @@ If `pi05_piper_stacking` is not yet in your `src/openpi/training/config.py`, add
 Fine-tuned from Physical Intelligence's `pi05_base` checkpoint via openpi; the weights inherit the
 upstream openpi / base-model license. Dataset: axiboai/piper_stacking.
 
-
 ## Checkpoints in this repo
 
-Two checkpoints from the **same run** are provided so you can compare for overfitting:
+Checkpoints from the **same run** (for overfitting comparison):
 
-| Path | Step | Train loss | Load with |
-|---|---|---|---|
-| `/` (root) | 10000 (final) | 0.0012 | `create_trained_policy(cfg, ckpt_dir)` |
-| `step_2500/` | 2500 | 0.0048 | `create_trained_policy(cfg, ckpt_dir / "step_2500")` |
+| Path | Step | Train loss |
+|---|---|---|
+| `/` (root) | 10,000 (final) | 0.0012 |
+| `step_7500/` | 7,500 | 0.0021 |
+| `step_5000/` | 5,000 | 0.0031 |
+| `step_2500/` | 2,500 | 0.0048 |
 
-Training loss fell monotonically (2500: 0.0048 → 5000: 0.0031 → 7500: 0.0021 → final: 0.0012). Train loss alone does not reveal overfitting — compare **real-rollout success rate** of `step_2500/` vs the final to check whether the longer run over-memorized the 60 demos.
+Load any with `create_trained_policy(cfg, ckpt_dir / "step_XXXX")` (root = final). Training loss fell monotonically (2,500 → final); compare **real-rollout success** to tell whether the longer run overfit the 60 demos.