|
--- |
|
language: |
|
- el |
|
tags: |
|
- text |
|
- language-modeling |
|
datasets: |
|
- dataset/wiki_oscar_combined_normalized_uncased |
|
metrics: |
|
- accuracy |
|
model-index: |
|
- name: greek-longformer-base-4096 |
|
results: |
|
- task: |
|
name: Masked Language Modeling |
|
type: fill-mask |
|
dataset: |
|
name: dataset/wiki_oscar_combined_normalized_uncased |
|
type: dataset/wiki_oscar_combined_normalized_uncased |
|
split: None |
|
metrics: |
|
- name: Accuracy |
|
type: accuracy |
|
value: 0.7765486725663717 |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# Greek Longformer |
|
|
|
A Greek version of the Longformer Language Model. |
|
|
|
This model is a (from scratch) Greek Longformer model based on the configuration of [allenai/longformer-base-4096](https://huggingface.co/allenai/longformer-base-4096), and trained on the combined datasets from the [Greek Wikipedia](https://huggingface.co/datasets/wikipedia) and the Greek part of [OSCAR](https://huggingface.co/datasets/oscar-corpus/OSCAR-2301). |
|
It achieves the following results on the evaluation set: |
|
|
|
- Loss: 1.1080 |
|
- Accuracy: 0.7765 |
|
|
|
## Pre-training corpora |
|
|
|
The pre-training corpora of `greek-longformer-base-4096` include: |
|
|
|
- The Greek part of [Wikipedia](https://el.wikipedia.org/wiki/Βικιπαίδεια:Αντίγραφα_της_βάσης_δεδομένων), |
|
- The Greek part of [OSCAR](https://traces1.inria.fr/oscar/), a cleansed version of [Common Crawl](https://commoncrawl.org). |
|
|
|
## Model description |
|
|
|
More information needed |
|
|
|
## Intended uses & limitations |
|
|
|
More information needed |
|
|
|
## Training and evaluation data |
|
|
|
More information needed |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
|
|
- learning_rate: 5e-05 |
|
- train_batch_size: 8 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 8 |
|
- total_train_batch_size: 64 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- lr_scheduler_warmup_ratio: 0.1 |
|
- num_epochs: 6.0 |
|
|
|
### Training results |
|
|
|
### Framework versions |
|
|
|
- Transformers 4.28.0.dev0 |
|
- Pytorch 2.0.0+cu118 |
|
- Datasets 2.11.0 |
|
- Tokenizers 0.13.2 |
|
|
|
## Citing & Authors |
|
|
|
The model has been officially released with the article "From Pre-training to Meta-Learning: A journey in Low-Resource-Language Representation Learning". |
|
Dimitrios Zaikis and Ioannis Vlahavas. |
|
In: IEEE Access. |
|
|
|
If you use the model, please cite the following: |
|
|
|
```bibtex |
|
|
|
@ARTICLE{10288436, |
|
author = {Zaikis, Dimitrios and Vlahavas, Ioannis}, |
|
journal = {IEEE Access}, |
|
title = {From Pre-training to Meta-Learning: A journey in Low-Resource-Language Representation Learning}, |
|
year = {2023}, |
|
volume = {}, |
|
number = {}, |
|
pages = {1-1}, |
|
doi = {10.1109/ACCESS.2023.3326337} |
|
} |
|
|
|
``` |
|
|