t5-eng-fra

---

library_name: transformers license: apache-2.0 base_model: t5-small tags:

• generated_from_trainer
• translation
• t5
• english-to-french model-index:
• name: t5-fra-eng results:
  • task: translation dataset: Custom English-French dataset https://huggingface.co/datasets/SOULAMA/eng-fra-dataset

---

<!-- 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. -->

t5-eng-fra

This model is a fine-tuned version of t5-small on a French-English translation dataset.

Model description

t5-eng-fra is a sequence-to-sequence model designed to translate English text into French, and can be use to to translate French-English with the same dataset. It was fine-tuned from t5-small using Hugging Face’s Seq2SeqTrainer on a custom French-English dataset. The model leverages T5’s transformer architecture and the text-to-text paradigm for translation tasks.

Intended uses & limitations

Intended uses:

• Automatic translation of English text into French.
• Machine translation research and experimentation.

Limitations:

• The model may produce incorrect translations for idiomatic expressions or rare vocabulary.
• Not suitable for legal, medical, or critical translations without human verification.
• Performance depends on the quality and size of the fine-tuning dataset.

Training and evaluation data

• Dataset: Custom French-English parallel sentences.
• Split: 80% training, 20% test from dataset.
• Split 80% training, 20% test from trainset
• Data preprocessing: Text normalized, tokenized using t5-small tokenizer, maximum input length = 128 tokens, maximum target length = 128 tokens.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-5
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 3
• weight_decay: 0.01
• optimizer: AdamW (betas=(0.9, 0.999), epsilon=1e-8)
• lr_scheduler_type: linear
• mixed_precision_training: Native AMP
• seed: 42
• evaluation_strategy: epoch
• save_total_limit: 3
• predict_with_generate: True

Training results

• 'eval_loss': 0.5946913957595825,
• 'eval_bleu': 42.4875053753255,
• 'eval_chrf': 61.82547115972855,}

Framework versions

• Transformers 4.57.3
• PyTorch 2.6.0+cu126
• Datasets 3.6.0
• Tokenizers 0.22.1