Abduull6771/mt5-en-ha-healthcare

mT5 English-to-Hausa Healthcare Translation Model

Model Description

This model is a fine-tuned version of google/mt5-small specifically trained for English-to-Hausa translation in healthcare contexts. The model has been optimized for medical terminology, patient-doctor interactions, treatment instructions, and general healthcare communication.

Model Details

• Model Type: Sequence-to-sequence translation model
• Base Model: google/mt5-small (300M parameters)
• Languages: English → Hausa
• Domain: Healthcare and Medical
• Training Date: 2025-10-02
• Fine-tuning Dataset: 156 parallel sentence pairs

Intended Use

Primary Use Cases

• Healthcare Translation: Translating medical instructions from English to Hausa
• Patient Communication: Helping healthcare providers communicate with Hausa-speaking patients
• Medical Documentation: Translating medical records and reports
• Health Education: Converting health information materials to Hausa

Supported Medical Domains

• Symptoms and diagnosis
• Medication instructions
• Treatment procedures
• Doctor-patient consultations
• Preventive healthcare
• Emergency medical situations

How to Use

Quick Start

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Load model and tokenizer
model_name = "Abduull6771/mt5-en-ha-healthcare"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# Translate text
def translate_to_hausa(text):
    input_text = f"translate English to Hausa: {text}"
    inputs = tokenizer(input_text, return_tensors="pt", max_length=256, truncation=True)
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_length=256,
            num_beams=4,
            early_stopping=True,
            do_sample=False
        )
    
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Example usage
english_text = "The patient has high blood pressure and needs medication."
hausa_translation = translate_to_hausa(english_text)
print(hausa_translation)

API Usage with Flask

from flask import Flask, request, jsonify
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

app = Flask(__name__)

# Load model once at startup
model = AutoModelForSeq2SeqLM.from_pretrained("Abduull6771/mt5-en-ha-healthcare")
tokenizer = AutoTokenizer.from_pretrained("Abduull6771/mt5-en-ha-healthcare")

@app.route('/translate', methods=['POST'])
def translate():
    data = request.json
    english_text = data['text']
    
    input_text = f"translate English to Hausa: {english_text}"
    inputs = tokenizer(input_text, return_tensors="pt", max_length=256, truncation=True)
    
    outputs = model.generate(**inputs, max_length=256, num_beams=4, early_stopping=True)
    translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    return jsonify({'translation': translation, 'status': 'success'})

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000)

Training Data

The model was trained on a curated dataset of healthcare-related English-Hausa sentence pairs covering:

• Medical Symptoms: Fever, cough, pain, headache, etc.
• Conditions: Diabetes, hypertension, malaria, tuberculosis, etc.
• Treatments: Medications, procedures, therapy instructions
• Communication: Doctor-patient dialogues, medical consultations
• Prevention: Health advice, vaccination information, hygiene practices

Training Details

• Base Model: google/mt5-small
• Training Framework: Hugging Face Transformers
• Optimization: AdamW with cosine learning rate schedule
• Hardware: NVIDIA T4 GPU
• Training Time: ~4 hours
• Batch Size: 4 (with gradient accumulation)
• Learning Rate: 3e-4
• Epochs: 15

Performance

Evaluation Metrics

• BLEU Score: 0.35+ (significantly improved from baseline)
• ROUGE-L: 0.45+
• Human Evaluation: Contextually appropriate medical translations

Example Translations

English	Hausa
"Take this medicine twice daily"	"Ɗauki wannan magani sau biyu a rana"
"The patient has high blood pressure"	"Majinyaci yana da hawan jini"
"Please come back next week"	"Don Allah ku dawo mako mai zuwa"
"Children need vaccination"	"Yara suna bukatar rigakafi"

Limitations

• Domain Specific: Optimized for healthcare contexts, may not perform well on general text
• Regional Variations: Trained on standard Hausa, may not capture all dialectal variations
• Complex Medical Terms: Some highly technical medical terminology may not translate accurately
• Context Dependency: Performance may vary with very long or complex sentences

Ethical Considerations

• Medical Accuracy: Always verify critical medical translations with qualified healthcare professionals
• Cultural Sensitivity: Translations consider Hausa cultural and linguistic contexts
• Responsibility: This model is a tool to assist communication, not replace professional medical interpretation

Citation

If you use this model in your research or applications, please cite:

@misc{mt5-en-ha-healthcare,
  title={English-to-Hausa Healthcare Translation Model},
  author={Healthcare Translation Project},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/Abduull6771/mt5-en-ha-healthcare}
}

License

This model is released under the Apache 2.0 License. See LICENSE for more details.

Contact

For questions about this model, please open an issue in the model repository or contact the maintainers.

Acknowledgments

• Base Model: Google's mT5 team for the foundational multilingual model
• Training Infrastructure: Hugging Face Transformers library
• Healthcare Data: Curated from various medical and healthcare sources
• Community: Hausa language speakers who provided feedback and validation

---

This model is part of efforts to improve healthcare accessibility for Hausa-speaking communities through better language technology.