These results confirm that the adapter-training pipeline completed successfully. They do not prove that the adapter is more accurate than the base model across all tasks.

Preliminary evaluation

Initial testing showed mixed results:

Business-writing responses were generally clear and professional.
Customer-service responses were polite but sometimes omitted requested details.
Nigerian Pidgin translation quality was inconsistent.
Some terminology responses contained factual errors.
The model occasionally added information that was not present in the prompt.

This release should therefore be treated as a research prototype.

Limitations

The model may:

Produce inaccurate information
Change or omit details during translation
Generate unnatural Nigerian Pidgin
Misunderstand cultural or business context
Add unsupported facts or explanations
Hallucinate business, legal, tax, financial, or regulatory guidance

Human review is required before real-world use.

Loading the adapter

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "Qwen/Qwen2.5-0.5B-Instruct"
adapter_id = "mgbam/gaialab-naija-adapter-v0.1"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    dtype=torch.float16,
    device_map="auto"
)

model = PeftModel.from_pretrained(
    base_model,
    adapter_id
)

model.eval()

Load the tokenizer from the base model because it contains the correct
Qwen chat template.

Model repository

Hugging Face:

mgbam/gaialab-naija-adapter-v0.1

Source code

GitHub:

oluwafemidiakhoa/gaialab-naija-assistant

Disclaimer

This is an experimental research release. Generated responses should be
reviewed before use in business, legal, financial, tax, medical, or
regulatory contexts.