text_generation_bangla_model

BanglaCLM dataset:

OSCAR: 12.84GB
Wikipedia dump: 6.24GB
ProthomAlo: 3.92GB
Kalerkantho: 3.24GB

Model description

context size : 128

Training and evaluation data

The BanglaCLM data set is divided into a training set (90%)and a validation set (10%).

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

Batch size: 32
Initial learning rate: 5e-5
Number of warmup steps: 10000
Weight decay rate: 0.01
Tokenization algorithm: BPE
Vocabulary size of tokenizer: 50256
Total trainable params: 124,439,808
Epochs: 40
Number of training steps: 40772228
training_precision: float32

Training results

perplexity score: 2.86.

Framework versions

Transformers 4.26.1
TensorFlow 2.11.0
Datasets 2.10.0
Tokenizers 0.13.2

Citation

If you find this model helpful, please cite.

@INPROCEEDINGS{10303383,
  author={Salim, Md. Shahidul and Murad, Hasan and Das, Dola and Ahmed, Faisal},
  booktitle={2023 International Conference on Information and Communication Technology for Sustainable Development (ICICT4SD)}, 
  title={BanglaGPT: A Generative Pretrained Transformer-Based Model for Bangla Language}, 
  year={2023},
  volume={},
  number={},
  pages={56-59},
  doi={10.1109/ICICT4SD59951.2023.10303383}}