AraGPT2
You can find more information in our paper AraGPT2
The code in this repository was used to train all GPT2 variants. The code support training and fine-tuning GPT2 on GPUs and TPUs via the TPUEstimator API.
GPT2-base and medium uses the code from the gpt2 folder and can trains models from the minimaxir/gpt-2-simple repository.
These models were trained using the lamb optimizer and follow the same architecture as gpt2 and are fully compatible with the transformers library.
GPT2-large and GPT2-mega were trained using the imcaspar/gpt2-ml library, and follow the grover architecture. You can use the pytorch classes found in grover/modeling_gpt2.py as a direct replacement for classes in the transformers library (it should support version v4.x from transformers).
Both models are trained using the adafactor optimizer, since the adam and lamb optimizer use too much memory causing the model to not even fit 1 batch on a TPU core.
AraGPT2 is trained on the same large Arabic Dataset as AraBERTv2.
AraGPT2 Detector
Machine generated detector model from the AraGPT2: Pre-Trained Transformer for Arabic Language Generation paper This model is trained on the long text passages, and achieves a 99.4% F1-Score.
Usage
Testing the model using transformers:
from transformers import GPT2TokenizerFast, pipeline
#for base and medium
from transformers import GPT2LMHeadModel
#for large and mega
from arabert.aragpt2.grover.modeling_gpt2 import GPT2LMHeadModel
from arabert.preprocess import ArabertPreprocessor
MODEL_NAME='aubmindlab/aragpt2-base'
arabert_prep = ArabertPreprocessor(model_name=MODEL_NAME)
model = GPT2LMHeadModel.from_pretrained(MODEL_NAME)
tokenizer = GPT2TokenizerFast.from_pretrained(MODEL_NAME)
generation_pipeline = pipeline("text-generation",model=model,tokenizer=tokenizer)
text=""
text_clean = arabert_prep.preprocess(text)
#feel free to try different decoding settings
generation_pipeline(
text_clean,
pad_token_id=tokenizer.eos_token_id,
num_beams=10,
max_length=200,
top_p=0.9,
repetition_penalty = 3.0,
no_repeat_ngram_size = 3)[0]['generated_text']
How to use the detector:
from transformers import pipeline
from arabert.preprocess import ArabertPreprocessor
processor = ArabertPreprocessor(model="aubmindlab/araelectra-base-discriminator")
pipe = pipeline("sentiment-analysis", model = "aubmindlab/aragpt2-mega-detector-long")
text = " "
text_prep = processor.preprocess(text)
result = pipe(text_prep)
# [{'label': 'machine-generated', 'score': 0.9977743625640869}]
Finetunning using transformers:
Follow the guide linked here
Finetuning using our code with TF 1.15.4:
- Create the Training TFRecords:
python create_pretraining_data.py
--input_file=<RAW TEXT FILE with documents/article sperated by an empty line>
--output_file=<OUTPUT TFRecord>
--tokenizer_dir=<Directory with the GPT2 Tokenizer files>
- Finetuning:
python3 run_pretraining.py \
--input_file="gs://<GS_BUCKET>/pretraining_data/*" \
--output_dir="gs://<GS_BUCKET>/pretraining_model/" \
--config_file="config/small_hparams.json" \
--batch_size=128 \
--eval_batch_size=8 \
--num_train_steps= \
--num_warmup_steps= \
--learning_rate= \
--save_checkpoints_steps= \
--max_seq_length=1024 \
--max_eval_steps= \
--optimizer="lamb" \
--iterations_per_loop=5000 \
--keep_checkpoint_max=10 \
--use_tpu=True \
--tpu_name=<TPU NAME> \
--do_train=True \
--do_eval=False
Model Sizes
| Model | Optimizer | Context size | Embedding Size | Num of heads | Num of layers | Model Size / Num of Params |
|---|---|---|---|---|---|---|
| AraGPT2-base | lamb |
1024 | 768 | 12 | 12 | 527MB / 135M |
| AraGPT2-medium | lamb |
1024 | 1024 | 16 | 24 | 1.4GB / 369M |
| AraGPT2-large | adafactor |
1024 | 1280 | 20 | 36 | 2.98GB/792M |
| AraGPT2-mega | adafactor |
1024 | 1536 | 24 | 48 | 5.5GB/1.46B |
Compute
| Model | Hardware | num of examples (seq len = 1024) | Batch Size | Num of Steps | Time (in days) |
|---|---|---|---|---|---|
| AraGPT2-base | TPUv3-128 | 9.7M | 1792 | 125K | 1.5 |
| AraGPT2-medium | TPUv3-8 | 9.7M | 80 | 1M | 15 |
| AraGPT2-large | TPUv3-128 | 9.7M | 256 | 220k | 3 |
| AraGPT2-mega | TPUv3-128 | 9.7M | 256 | 780K | 9 |
Results
The results show in the table below are the perplexity values on wikipedia articles that are not in the training data.
| Model | PPL |
|---|---|
| AraGPT2-base | 55.8 |
| AraGPT2-medium | 45.7 |
| AraGPT2-large | 36.6 |
| AraGPT2-mega | 29.8 |
Disclaimer
The text generated by AraGPT2 Arabic is automatically generated by a neural network model trained on a large amount of texts, which does not represent the authors' or their institutes' official attitudes and preferences. The text generated by AraGPT2 should only be used for research and scientific purposes. If it infringes on your rights and interests or violates social morality, please do not propagate it.
If you used this model please cite us as :
@inproceedings{antoun-etal-2021-aragpt2,
title = "{A}ra{GPT}2: Pre-Trained Transformer for {A}rabic Language Generation",
author = "Antoun, Wissam and
Baly, Fady and
Hajj, Hazem",
booktitle = "Proceedings of the Sixth Arabic Natural Language Processing Workshop",
month = apr,
year = "2021",
address = "Kyiv, Ukraine (Virtual)",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2021.wanlp-1.21",
pages = "196--207",
}
Acknowledgments
Thanks to TensorFlow Research Cloud (TFRC) for the free access to Cloud TPUs, couldn't have done it without this program, and to the AUB MIND Lab Members for the continous support. Also thanks to Yakshof and Assafir for data and storage access. Another thanks for Habib Rahal (https://www.behance.net/rahalhabib), for putting a face to AraBERT.
Contacts
Wissam Antoun: Linkedin | Twitter | Github | wfa07@mail.aub.edu | wissam.antoun@gmail.com