---
library_name: transformers
language:
- nep
- hi
- sa
- mr
base_model: RoBERTa
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
- precision
- recall
model-index:
- name: RoBERTa-devangari-script-classification
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# RoBERTa-devangari-script-classification

This model is a fine-tuned version of [RoBERTa](https://huggingface.co/RoBERTa) on the Custom Devangari Datasets dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0329
- Accuracy: 0.9935
- F1: 0.9935
- Precision: 0.9935
- Recall: 0.9935

## Model description

This model is a fine-tuned version of RoBERTa, optimized for multiclass text classification on datasets written in 
Devanagari script across multiple languages, including Nepali, Marathi, Sanskrit, Bhojpuri, and Hindi. By leveraging the 
robust RoBERTa architecture, this model has been fine-tuned to recognize intricate patterns and contextual 
cues within Devanagari text, achieving high accuracy and F1 scores for multiclass classification tasks.

## Intended uses & limitations

#### Intended Uses:

- Multiclass text classification for Nepali, Marathi, Sanskrit, Bhojpuri, and Hindi, written in Devanagari script.
- Suitable for sentiment analysis, topic categorization, and public opinion monitoring.

#### Limitations:

- Limited to Devanagari script; accuracy may drop on other scripts.
- Fine-tuned for multiclass classification; may not generalize well to other tasks or binary classifications.
- Language-specific nuances not present in the dataset may impact performance on certain dialects.

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch  | Step | Validation Loss | Accuracy | F1     | Precision | Recall |
|:-------------:|:------:|:----:|:---------------:|:--------:|:------:|:---------:|:------:|
| 0.2337        | 1.0 | 1638 | 0.0603          | 0.9874   | 0.9874 | 0.9875    | 0.9874 |
| 0.0513        | 2.0    | 3277 | 0.0387          | 0.9919   | 0.9919 | 0.9919    | 0.9919 |
| 0.0252        | 3.0 | 4914 | 0.0329          | 0.9935   | 0.9935 | 0.9935    | 0.9935 |

### Framework versions

- Transformers 4.44.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.19.1