Instructions to use Dataphyte/nubia_ai with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Dataphyte/nubia_ai with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B") model = PeftModel.from_pretrained(base_model, "Dataphyte/nubia_ai") - Notebooks
- Google Colab
- Kaggle
Nubia AI
Description (English)
Nubia AI, built by Dataphyte, is an AI assistant optimized for African languages, with dedicated support for agriculture and healthcare. It currently answers in Yoruba and English, and its strongest area is practical farming advice, such as planting, caring for crops and keeping livestock healthy. Health support and more African languages are being developed. It is built on the open-source Qwen3.8-27B model and trained further on Yoruba farming questions and answers.
Results so far
The figures below were measured on the earlier version of this adapter, before the name was added, on 500 unseen Yoruba farming questions. They compare how closely answers match a reference answer (chrF++, 0 to 100, higher is better); they do not show that the advice is correct.
| Before Yoruba training | After | |
|---|---|---|
| chrF++ | 14.58 | 21.85 |
| BLEU | 1.69 | 5.58 |
The trained answer was closer to the reference on 446 of 500 questions. These tests should be repeated on this version.
Limits
- Answers are not yet reviewed by agriculture experts.
- It has been trained on Yoruba farming data only so far. Health support and other African languages are still being developed, and it is not a medical or legal adviser.
- Answers are shorter than the reference answers on average.
Licence and credit
Built by Dataphyte on Qwen3.8-27B (Apache 2.0). This adapter modifies that model; credit to the Qwen team is retained.
- Downloads last month
- 45
Model tree for Dataphyte/nubia_ai
Base model
Qwen/Qwen3.8-27B