Downloads · 30 days
214
1% of all-time downloads
lighteternal/gpt2-finetuned-greek
gpt2-finetuned-greek is a text generation model from lighteternal. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
<img src="https://huggingface.co/lighteternal/gpt2-finetuned-greek-small/raw/main/GPT2el.png" width="600"/
Downloads · 30 days
214
1% of all-time downloads
All-time downloads
23.6K
Public
Repo size
1.5 GB
Likes
9
Public
Click a slice to open those files.
.bin510 MB · 50%
From the Hugging Face model README
A text generation (autoregressive) model, using Huggingface transformers and fastai based on the English GPT-2.
Finetuned with gradual layer unfreezing. This is a more efficient and sustainable alternative compared to training from scratch, especially for low-resource languages.
Based on the work of Thomas Dehaene (ML6) for the creation of a Dutch GPT2: https://colab.research.google.com/drive/1Y31tjMkB8TqKKFlZ5OJ9fcMp3p8suvs4?usp=sharing
from transformers import pipeline
model = "lighteternal/gpt2-finetuned-greek"
generator = pipeline(
'text-generation',
device=0,
model=f'{model}',
tokenizer=f'{model}')
text = "Μια φορά κι έναν καιρό"
print("\
".join([x.get("generated_text") for x in generator(
text,
max_length=len(text.split(" "))+15,
do_sample=True,
top_k=50,
repetition_penalty = 1.2,
add_special_tokens=False,
num_return_sequences=5,
temperature=0.95,
top_p=0.95)]))
We used a 23.4GB sample from a consolidated Greek corpus from CC100, Wikimatrix, Tatoeba, Books, SETIMES and GlobalVoices containing long senquences. This is a better version of our GPT-2 small model (https://huggingface.co/lighteternal/gpt2-finetuned-greek-small)
| Metric | Value |
|---|---|
| Train Loss | 3.67 |
| Validation Loss | 3.83 |
| Perplexity | 39.12 |
The research work was supported by the Hellenic Foundation for Research and Innovation (HFRI) under the HFRI PhD Fellowship grant (Fellowship Number:50, 2nd call)
Based on the work of Thomas Dehaene (ML6): https://blog.ml6.eu/dutch-gpt2-autoregressive-language-modelling-on-a-budget-cff3942dd020