mclanorjeff/prn-nurse-maternal-health-mt5-twi-english

Ghana Maternal Health mT5 Twi-English

This model is a fine-tuned google/mt5-small text-to-text model for Ghana maternal-health question answering in a Twi-English code-mixed setting.

Intended Use

The model is intended for educational maternal-health support and prototype development. It can be used to generate English maternal-health responses from English or Twi-English code-mixed questions.

Recommended workflow:

1. 1.Check for urgent danger signs.
2. 2.Retrieve a verified maternal-health answer from a curated Q&A source.
3. 3.Optionally use this model to rephrase the response.
4. 4.Tell the user to consult a nurse, midwife, doctor, or health facility when needed.

Training Data

Dataset: Ghana Maternal Health Twi-English Q&A Dataset

Dataset structure:

• —question: Twi-English code-mixed maternal-health question
• —question_english: English version of the question
• —answer: English answer
• —metadata: topic, stage, setting, urgency, and keywords

The model was trained on the code-mixed question as input and the English answer as output.

Training Summary

Metric progression:

Usage

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch

model_id = "mclanorjeff/prn-nurse-maternal-health-mt5-twi-english"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

question = "When should I start breastfeeding after delivery?"
prompt = (
    "Answer this maternal health question clearly and safely. "
    "Language: twi_english_code_mixed.\n"
    f"Question: {question}\n"
    "Answer:"
)

inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=256).to(device)
outputs = model.generate(**inputs, max_new_tokens=160, num_beams=4)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Example Prompt

Answer this maternal health question clearly and safely. Language: twi_english_code_mixed.
Question: Me ho yɛ me sɛ mebɛfe every morning. Is it normal?
Answer:

Gradio Space

A Gradio demo can load this model with transformers and expose a simple question-answer interface. The recommended demo includes a safety pre-check for danger signs before model generation.

Limitations and Safety

This model is not a doctor, nurse, midwife, or diagnostic system. It should not be used as a standalone medical chatbot.

Local testing showed that the model can sometimes produce fluent but wrong-topic answers. For safer real-world use, combine it with:

• —retrieval from verified Q&A records,
• —danger-sign detection,
• —human clinical review,
• —clear advice to seek professional care.

Urgent symptoms such as bleeding, severe headache, fever, convulsions, severe abdominal pain, trouble breathing, swollen face/hands, or reduced baby movement require immediate professional care. In Ghana, emergency help may be available via 112 or 193.