gha-understand-twi-medical-plus-language-v3

teckedd/gha-understand-twi-medical-plus-language-v3

Twi/Akan semantic-recovery LoRA for Ghana Health AI. Trained on a cleaned medical plus language-coverage silver corpus for research evaluation.

• Base model: Qwen/Qwen2.5-1.5B-Instruct
• Task: text-generation
• Languages: tw, ak, en
• Card generated: 2026-09-04 (UTC)
• Product: ghanahealth.serendepify.com

Not a medical device. Outputs support community health guidance only.

Metrics

Training data

• ghananlpcommunity/ghana-health-symptoms (cc-by-nc-4.0)
• google/WaxalNLP (language coverage rows)

Intended use

• In-product ASR / TTS / chat for Twi (Akan) and English health conversations in Ghana.
• Research on low-resource Ghanaian language speech and health dialogue.

Out of scope

• Clinical diagnosis or autonomous medical decisions.
• Claiming near-native quality without reporting held-out WER/CER or human A/B scores.

How to load

from transformers import pipeline  # or AutoModel + processor per task
# repo: teckedd/gha-understand-twi-medical-plus-language-v3

Citation

If you use this checkpoint, please credit Ghana Health AI / Serendepify and the upstream base model authors plus any listed datasets.

Dataset status

This is a research checkpoint trained from machine annotations. Rows are not human-gold labels. The main medical source is CC-BY-NC-4.0, so use is non-commercial research unless separate permission is obtained.

Source notes

• Ghana Health Symptoms rows are the primary medical semantic-recovery source.
• WAXAL rows are included only as language-coverage silver rows after filtering.
• GhanaNLP speech-text rows are included as local language-coverage silver rows; they are described in the card body, not HF front-matter, until the source id and license are audited.
• Product-failure seed rows cover observed app failures such as eye pain, child fever, hospital navigation, malaria follow-ups, and Twi commerce purchase/search requests.

Training run

• Train rows: 6329
• Steps: 650
• Corpus: data/understanding-corpus/silver-medical-plus-language-v1
• Final train loss: 0.7946

Product fixture evaluation

• Status: needs_review
• Passed: 7/11
• Pass rate: 63.64%
• Artifact: eval/product-fixtures.v0.json
• Production note: this checkpoint is exposed only as an explicit research option.