NeoDictaBERT: Pushing the Frontier of BERT models in Hebrew
Following the success of
ModernBERT
and
NeoBERT
, we set out to train a Hebrew version of NeoBERT.
Introducing
NeoDictaBERT
: A Next-Generation BERT-style model trained specifically for Hebrew, technical report coming soon.
Supported Context Length:
4,096
(~
3,200
words)
Trained on a total of 235B tokens (5 epochs) with a context length of 1,024, and another 50B tokens with a context length of 4,096.
This is the base model pretrained with the masked-language-modeling objective.
Sample usage:
from transformers import AutoModelForMaskedLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('dicta-il/neodictabert')
model = AutoModelForMaskedLM.from_pretrained('dicta-il/neodictabert', trust_remote_code=True)
model.eval()
sentence = 'בשנת 1948 השלים אפרים קישון את [MASK] בפיסול מתכת ובתולדות האמנות והחל לפרסם מאמרים הומוריסטיים'
output = model(tokenizer.encode(sentence, return_tensors='pt'))
# the [MASK] is the 7th token (including [CLS])import torch
top_2 = torch.topk(output.logits[0, 7, :], 2)[1]
print('\n'.join(tokenizer.convert_ids_to_tokens(top_2))) # should print לימודיו / הכשרתו
Performance
Model
HeQ EM
HeQ F1
HeQ TInls
Syntax-HTBv2=>HTBv2 UAS
Syntax-HTBv2=>HTBv2 LAS
Sentiment Acc
NER (NEMO) F1
Segmentation Acc
mBERT
69.08
74.32
77.01
--
--
84.21
79.11
--
AlephBertGimmel
61.77
67.97
71.35
--
--
89.51
85.26
--
DictaBERT
70.35
77.04
80.34
90.52
86.54
89.79
87.01
99.1
NeoDictaBERT_1024
75.07
82.20
85.51
91.48
87.21
--
--
--
NeoDictaBERT_4096
76.40
82.83
85.86
91.77
87.80
89.61
86.97
99.2
Δ vs DictaBERT
+6.05
+5.79
+5.52
+1.25
+1.26
-0.18
-0.04
+0.1
Citation
If you use NeoDictaBERT in your research, please cite
NeoDictaBERT: Pushing the Frontier of BERT models in Hebrew
BibTeX:
@misc{NeoDictaBERT,
title = {NeoDictaBERT: Pushing the Frontier of BERT models in Hebrew},
author = {Shaltiel Shmidman and Avi Shmidman and Moshe Koppel},
howpublished = {Hugging Face},
url = {https://huggingface.co/dicta-il/neodictabert},
license = {CC BY 4.0},
note = {Technical report coming soon}
}
neodictabert huggingface.co is an AI model on huggingface.co that provides neodictabert's model effect (), which can be used instantly with this dicta-il neodictabert model. huggingface.co supports a free trial of the neodictabert model, and also provides paid use of the neodictabert. Support call neodictabert model through api, including Node.js, Python, http.
neodictabert huggingface.co is an online trial and call api platform, which integrates neodictabert's modeling effects, including api services, and provides a free online trial of neodictabert, you can try neodictabert online for free by clicking the link below.
dicta-il neodictabert online free url in huggingface.co:
neodictabert is an open source model from GitHub that offers a free installation service, and any user can find neodictabert on GitHub to install. At the same time, huggingface.co provides the effect of neodictabert install, users can directly use neodictabert installed effect in huggingface.co for debugging and trial. It also supports api for free installation.