KDE4_Dataset_Translation_English_to_Hindi_with_opus_mt_en_hi

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-hi on the kde4 dataset.

Model description

Helsinki-NLP/opus-mt-en-hi

Intended uses & limitations

English to Hindi Translation

Training and evaluation data

KDE4 dataset english to hindi, available at: KDE4-en2hi

Since most sentences are small and contain at most around 12 words, we truncate all sentences to 20 tokens during tokenization.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Before training, BLEU score on the validation set: 59.886508891144366,

After training, BLEU score on the validation set: 54.98374943358887

As we reduce the max_length truncation parameter during tokenization, the BLEU score after training improves.

Framework versions

• Transformers 4.50.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1