devesh-002 / Major_Project_IREL

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

IRE Project: Abstractive Text Summarisation

  1. MCLAS
    Code: [Link to mT5 repo] Paper

  2. mT5
    Code: [Link to repo]

Training vanilla mT5. Currently it is in Gujarati: Training of mT5 (The Colab files have the log data).

Predicting summary by mT5. Predicting mT5

This contains fine-tuned Gujarati Summarisation model (trained up to 3 epochs).

Paper

Fine-tuned code for mT5. Fine-tuned mT5

Datasets used:

Gujarati

Hindi and English, ignore or replace gujarati folder in this.

  1. IndicBART Code

Results:

Vanilla mT5:

Language Rouge-1 Rouge-2 Rouge-L
English 48.6645 36.1859 43.69
Hindi 51.468 40.1589 46.6524
Gujarati 23.0882 14.0092 20.7578

XL-Sum:

Language Rouge-1 Rouge-L
English 43.35 34.47
Hindi 41.79 36.67

Contributors:
Aaradhya Gupta
Devesh Marwah
Mayank Goel
Radheshyam Thiyagarajan

About


Languages

Language:Python 100.0%