T-XLM-RoBERTa-Hindi-Sentiment model is a fine-tuned version of the Twitter-XLM-RoBERTa-base model from Cardiff-NLP.

txlm-roberta-hindi-sentiment is a Hindi language sentiment classifier model (in Devanagari script) which is trained on a publicly available Hindi language dataset. See the GitHub source of the dataset HERE.

The training, testing and validation datasets consist of 6807, 1634 and 635 numbers of labelled Hindi language examples respectively.

The trained model shows a weighted average macro F1-score of 0.89 (please see the confusion matrix in the Google Colab notebook below).

The Google Colab notebook, where the model is fine-tuned by employing native PyTorch modules can be found on LondonStory's GitHub page HERE.