Card: point to the newer D version on GitHub
Browse files
README.md
CHANGED
|
@@ -30,6 +30,12 @@ file that keeps almost all of the float model's accuracy.
|
|
| 30 |
This replaces [inviolus/erdem-ocr-tr-tinyml](https://huggingface.co/inviolus/erdem-ocr-tr-tinyml)
|
| 31 |
(2026-10-09), whose int8 file lost most of its accuracy.
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
## Files
|
| 34 |
|
| 35 |
| File | What it is |
|
|
|
|
| 30 |
This replaces [inviolus/erdem-ocr-tr-tinyml](https://huggingface.co/inviolus/erdem-ocr-tr-tinyml)
|
| 31 |
(2026-10-09), whose int8 file lost most of its accuracy.
|
| 32 |
|
| 33 |
+
> **Newer version on GitHub (2026-11-01):** a continuation of this model, *D*,
|
| 34 |
+
> trained with 47,613 extra printed-document lines (donut 200k, CC BY-SA 4.0),
|
| 35 |
+
> reads 46.5% of the donut test rows exactly as int8 (this model: 9.3%), with
|
| 36 |
+
> the same size and 1 to 2 points less on the sets below. The files here are
|
| 37 |
+
> still the 2026-10-31 model. Scores: [InviolusOCR results](https://github.com/ErdemWilkinson/InviolusOCR#2026-11-01-d-b-plus-the-donut-200k-lines).
|
| 38 |
+
|
| 39 |
## Files
|
| 40 |
|
| 41 |
| File | What it is |
|