---
base_model: Qwen/Qwen2.5-0.5B-Instruct
library_name: peft
pipeline_tag: text-generation
language:
  - en
tags:
  - lora
  - peft
  - nigeria
  - nigerian-english
  - nigerian-pidgin
  - customer-service
  - scam-safety
  - business-writing
license: apache-2.0
---

# GaiaLab Naija Assistant v0.5

GaiaLab Naija Assistant v0.5 is an experimental LoRA adapter for `Qwen/Qwen2.5-0.5B-Instruct`.

This release adds a reproducible dataset workflow for CSV ingestion, JSONL generation, validation, duplicate checking, statistics, and CPU-compatible LoRA training.

## Model Details

| Field | Value |
|---|---|
| Version | v0.5 |
| Base model | `Qwen/Qwen2.5-0.5B-Instruct` |
| Fine-tuning method | LoRA / PEFT |
| Model type | Causal language model adapter |
| Training examples | 47 |
| Dataset health score | 95/100 |
| Developer | Oluwafemi Idiakhoa |
| Project | GaiaLab AI |

## Training Categories

| Category | Examples |
|---|---:|
| Safety and scams | 13 |
| Professional boundaries | 12 |
| Customer service | 10 |
| Nigerian English | 10 |
| Business writing | 1 |
| Nigerian Pidgin | 1 |
| **Total** | **47** |

## Risk-Level Distribution

| Risk level | Examples |
|---|---:|
| High | 19 |
| Medium | 7 |
| Low | 21 |

## Dataset Validation

The v0.5 pipeline reported:

- Valid JSONL
- Required fields present
- Correct system, user, and assistant role order
- Zero duplicate IDs
- Zero duplicate prompts
- Zero missing prompts
- Zero missing responses
- Dataset health score of 95/100

## Evaluation Status

A formal side-by-side benchmark comparing v0.4 and v0.5 has not yet been published. This model card does not claim that v0.5 outperforms v0.4.

## Intended Uses

- Research and education
- Nigerian customer-service prototypes
- Professional message drafting
- Scam-awareness demonstrations
- Nigerian English experimentation
- Basic Nigerian Pidgin experimentation
- CPU-friendly LoRA research

## Limitations

- The training dataset contains only 47 examples
- Business writing and Pidgin each contain only one example
- The dataset is unevenly distributed
- The adapter may overfit specific wording
- Cultural coverage is narrow
- The model may hallucinate
- Human review is required for important outputs

## Installation

```bash
pip install torch transformers peft
```

## Usage

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

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

tokenizer = AutoTokenizer.from_pretrained(
    base_model_id,
    trust_remote_code=True,
)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float32,
    trust_remote_code=True,
)

model = PeftModel.from_pretrained(base_model, adapter_id)

messages = [
    {
        "role": "system",
        "content": (
            "You are GaiaLab Naija Assistant. Be helpful, concise, "
            "culturally aware, truthful, and safe."
        ),
    },
    {
        "role": "user",
        "content": "Write a polite payment reminder for a customer.",
    },
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=120,
        do_sample=False,
    )

new_tokens = output[0][inputs["input_ids"].shape[1]:]
print(tokenizer.decode(new_tokens, skip_special_tokens=True))
```

## Responsible Use

Do not use this model as the sole authority for medical, legal, financial, emergency, employment, identity-verification, or other high-impact decisions.

Never provide passwords, PINs, one-time passwords, bank verification codes, private keys, or other sensitive credentials to the model.

## Author

Developed by **Oluwafemi Idiakhoa** under the **GaiaLab AI** initiative.


## Project Links

- GitHub: https://github.com/oluwafemidiakhoa/gaialab-naija-assistant
- Model: https://huggingface.co/mgbam/gaialab-naija-adapter-v0.5
- GaiaLab AI: https://www.gailabai.com