metadata

language: si
tags:
  - Sinhala
  - text-generation
  - gpt2
datasets:
  - mc4

Sinhala GPT2 trained on MC4 (manually cleaned)

Overview

This is a smaller GPT2 model trained on MC4 Sinhala dataset. As Sinhala is one of those low resource languages, there are only a handful of models been trained. So, this would be a great place to start training for more downstream tasks.

This model uses a manually cleaned version of MC4 dataset which can be found here. Although the dataset is relatively small ~3GB. The finetuned model on news articles generates good and acceptable results.

Model Specification

The model chosen for training is GPT2 with the following specifications:

vocab_size=50257
n_embd=768
n_head=12
n_layer=12
n_positions=1024

How to Use

You can use this model directly with a pipeline for causal language modeling:

from transformers import pipeline
generator = pipeline('text-generation', model='flax-community/Sinhala-gpt2')
generator("මම", max_length=50, num_return_sequences=5)