palimpsest-greberta-continuation

BERT-base continued-pretrained from GreBerta (bowphs/GreBerta) on a new Ancient Greek corpus derived from the Harvard institutional-books collection.

Pretraining

  • Best val loss: 0.5385, PPL: 1.713 (epoch 40/40)

Downstream (full results in the paper/repo)

TAG POS MORPH LEMMA UAS LAS SENT_F
92.78 96.36 93.09 85.63 82.40 77.62 85.16
Downloads last month
18
Safetensors
Model size
0.1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for pnadel/palimpsest-greberta-continuation

Base model

bowphs/GreBerta
Finetuned
(8)
this model

Dataset used to train pnadel/palimpsest-greberta-continuation