Update README.md
Browse files
README.md
CHANGED
|
@@ -22,9 +22,6 @@ tags:
|
|
| 22 |
MEDUSA is a family of Vision Language Models (VLMs) fine-tuned for **multilingual medieval handwritten text recognition (HTR)** at the **line level**.
|
| 23 |
These models were developed at the [École nationale des chartes – PSL](https://www.chartes.psl.eu/).
|
| 24 |
|
| 25 |
-
This repository contains the `0.2` variants of MEDUSA, which extend the [0.1 release](https://huggingface.co/ENC-PSL/Medusa0.1Line-9B) with an explicit
|
| 26 |
-
**Silver pre-training stage** on synthetic line images rendered from text-only historical corpora.
|
| 27 |
-
|
| 28 |
> **Note on versioning.** The `0.1` variants correspond to the models submitted to the ICDAR 2026 competition, trained exclusively on Gold and Platinum image–text pairs.
|
| 29 |
> The `0.2` variants add a prior Silver stage using 500,000 synthetic lines, improving coverage of under-resourced languages.
|
| 30 |
|
|
|
|
| 22 |
MEDUSA is a family of Vision Language Models (VLMs) fine-tuned for **multilingual medieval handwritten text recognition (HTR)** at the **line level**.
|
| 23 |
These models were developed at the [École nationale des chartes – PSL](https://www.chartes.psl.eu/).
|
| 24 |
|
|
|
|
|
|
|
|
|
|
| 25 |
> **Note on versioning.** The `0.1` variants correspond to the models submitted to the ICDAR 2026 competition, trained exclusively on Gold and Platinum image–text pairs.
|
| 26 |
> The `0.2` variants add a prior Silver stage using 500,000 synthetic lines, improving coverage of under-resourced languages.
|
| 27 |
|