Archatext
/

hatformer-arams28k

vision-encoder-decoder

Model card Files Files and versions

HATFormer Fine-Tuned (AraMS-28k)

Historical Arabic Manuscript HTR model based on TrOCR (microsoft/trocr-large-stage1) fine-tuned on line-level annotations from the AraMS-28k dataset.

Downloads last month: -

Safetensors

Model size

0.3B params

Tensor type

F32

·

Model tree for Archatext/hatformer-arams28k

Base model

microsoft/trocr-large-stage1

Finetuned

(2)

this model