metadata

language:
  - hi
  - gu
  - mr
  - ml
  - si
  - bh
  - pa
  - bn
  - as
  - en

This model is the exclusive finetuning of 'Indo-Aryan-Abuse-Detection' model which was trained on approximately 1000 hindi labelled sentences.

Dataset details

Train set for Gujarati zero shot: HASOC-2023 Gujarati tweets labelled (200 samples) Test set: HASOC-2023 Gujarati unlabelled tweets (approximately 1000 samples)

Training details

32 batches 20 epochs

Results

Macro-F1: 0.7926 Macro precision: 0.7859