marian-finetuned-kde4-en-to-fr

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-fr on the kde4 dataset (a multilingual dataset for translation tasks).

Model description

A lightweight Marian-based model for English-to-French translation, fine-tuned for 3 epochs on the kde4 dataset. It achieves a BLEU score of 21.71 on the validation set, suitable for basic English-French translation scenarios (e.g., short texts, daily conversations).

Intended uses & limitations

• Small-scale English-to-French translation tasks (e.g., translating product descriptions, short articles).
• Educational purposes (learning how to fine-tune Marian translation models).
• May struggle with long texts (max sequence length: 128 tokens) or specialized domains (e.g., legal/medical terminology).

Training and evaluation data

• Training data: kde4 dataset (English-French subset), ~100k parallel sentences.
• Evaluation data: Random 10% subset of the kde4 dataset.

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_FUSED 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

Metric	Value
Eval Loss	2.307
BLEU Score	21.71

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1