Omarrran commited on
Commit
ce042f5
·
verified ·
1 Parent(s): ff3eda3

Document important OCR checkpoints

Browse files
Files changed (1) hide show
  1. CHECKPOINTS.md +20 -0
CHECKPOINTS.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Checkpoints
2
+
3
+ This repository intentionally publishes the important OCR checkpoints rather than the full 200-epoch training archive.
4
+
5
+ Recommended files:
6
+
7
+ - `checkpoints/best_accuracy.pdparams` — best validation checkpoint, epoch 191, selected by `norm_edit_dis`; recommended for inference/evaluation.
8
+ - `checkpoints/best_accuracy.pdopt` and `checkpoints/best_accuracy.states` — optimizer/state files for the best checkpoint when available.
9
+ - `checkpoints/latest.pdparams` — latest/final training checkpoint at epoch 200.
10
+ - `checkpoints/latest.pdopt` and `checkpoints/latest.states` — optimizer/state files for the latest checkpoint.
11
+ - `checkpoints/iter_epoch_191.pdparams` and `.states` — explicit best epoch checkpoint copy.
12
+ - `checkpoints/iter_epoch_200.pdparams` and `.states` — explicit final epoch checkpoint copy.
13
+
14
+ A duplicate copy of the same important files is also stored under:
15
+
16
+ `checkpoints/important/`
17
+
18
+ The full historical archive (`iter_epoch_1` through `iter_epoch_200` with every optimizer file) is not required for inference and is not treated as part of the minimal publishable release.
19
+
20
+ Dataset files are not included in this repository.