You can deploy this model and start earning money today!

Tushe/
AMINI-Culturally-Embedded-STEM-Hausa

Name: AMINI-Culturally-Embedded-STEM-Hausa
Availability: InStock

Views

No views yet

AMINI - Culturally-Embedded Hausa STEM Assistant

Model Description

AMINI (The Trustworthy One) is a fine-tuned Llama-3-8B-Instruct model specialized in providing STEM education in Hausa language with deep cultural embedding. Created by Tushe – The Foundry Research Team, AMINI bridges scientific excellence with Hausa cultural wisdom using the novel "Shehin Malamin Kimiyya" (The Way of the Science Teacher) framework.

Key Features

🌍 Culturally-Grounded: Explains STEM concepts using Hausa proverbs, analogies, and cultural contexts
🔬 STEM Mastery: Covers Physics, Chemistry, Biology, and Mathematics
🗣️ Pure Hausa: Uses descriptive Hausa phrases for technical terms, avoiding English-Hausa code-mixing
📚 Pedagogical Excellence: Teaches like a Hausa elder - patient, thorough, and engaging
🎯 Bilingual Input: Accepts questions in both English and Hausa, always responds in Hausa

Training Details

Base Model

Model: unsloth/llama-3-8b-Instruct-bnb-4bit
Parameters: 8 Billion
Quantization: 4-bit (bnb)

Fine-tuning Configuration

Method: LoRA (Low-Rank Adaptation)
LoRA Rank (r): 16
LoRA Alpha: 16
Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Training Framework: Unsloth + TRL SFTTrainer

Training Hyperparameters

Epochs: 3
Batch Size: 2 per device
Gradient Accumulation Steps: 4
Effective Batch Size: 8
Learning Rate: 2e-4
Optimizer: AdamW 8-bit
Scheduler: Linear
Max Sequence Length: 2048
Warmup Steps: 5

Dataset

Source: AlaminI/hausa-stem-reasoning-with-cultural-context
Size: 5,280 training examples (2,640 original samples × 2 formats)
Format: Instruction-following pairs (English→Hausa, Hausa→Hausa)
Domains: Physics, Chemistry, Biology, Mathematics

Training Infrastructure

GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition
Memory: 94.97 GB total, ~14 GB used
Training Time: ~35 minutes for 3 epochs
Total Steps: 1,980

The "Shehin Malamin Kimiyya" Framework

AMINI uses a structured teaching approach inspired by traditional Hausa pedagogy:

For Explanations:

KARIN MAGANA (Proverb) - Establish cultural framework
SHIMFIDA (Introduction) - Connect to familiar experiences
TSOKACI (Deep Dive) - Systematic explanation with analogies
HAKAI (Science Link) - Connect to formal scientific terminology
AIWATARWA (Application) - Practical uses in Hausa context

For Problem-Solving:

Fahimta (Understand) - Restate problem in cultural context
Tsara (Plan) - Outline solution approach using analogies
Aiki (Work) - Step-by-step solution with clear reasoning
Bincika (Check) - Verify solution makes sense
Bayani (Explain) - Connect solution to broader principles

Usage

Basic Usage

python

from unsloth import FastLanguageModel
import torch

# Load model
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="AlaminI/AMINI-Culturally-Embedded-STEM-Hausa",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True,
)

# Enable fast inference
FastLanguageModel.for_inference(model)

# Ask a question (in English or Hausa)
messages = [
    {"role": "user", "content": "Explain photosynthesis"}
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to("cuda")

# Generate response
outputs = model.generate(
    input_ids=inputs,
    max_new_tokens=512,
    temperature=0.7,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

With System Prompt (Recommended)

python

system_prompt = """You are AMINI - The Wise Scholar of Science, created by Tushe – The Foundry Research Team. You are a master of STEM subjects and a custodian of Hausa language and culture. Always respond in Hausa with cultural analogies."""

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": "Why is the sky blue?"}
]

# ... rest of the code

Example Outputs

Example 1: English Question → Hausa Response

Input: "Explain Newton's second law of motion"

Output:

Karin Magana: "Karfi yana haifar da motsi." (Force creates motion.)

Shimfida: Kamar yadda mai noma yake tura keken noma, haka karfi ke tura abubuwa...

Tsokaci: A kimiyyar motsi, muna da ka'ida mai suna Newton's Second Law. Wannan ka'ida tana cewa: Karfi (F) = Taro (m) × Hanzari (a), ko kuma F = ma...

[Full culturally-embedded explanation in Hausa]

Example 2: Hausa Question → Hausa Response

Input: "Me yasa jini yake ja?" (Why is blood red?)

Output:

Karin Magana:
"Jini shi ne rai."

Shimfida:
Idan mutum ya ji rauni kuma jini ya fito, za ka lura cewa launin sa ja ne sosai. Wannan ja mai karfi ba kwatsam ba ne; yana da dalilin kimiyya a jikinmu.

Tsokaci:
A cikin jini akwai wani sinadari mai suna haemoglobin. Wannan sinadari yana cikin kwayoyin jini ja, kuma aikinsa shi ne daukar iskar oxygen daga huhu zuwa sauran sassan jiki.

Haemoglobin yana dauke da wani sinadari mai suna iron (karfe). Lokacin da wannan karfe ya hadu da oxygen, sai ya ba jinin launi ja mai haske.

Misali:
Kamar yadda tsatsa a kan karfe yake sauya launin karfe, haka ma haduwar iron da oxygen a jini ke ba shi launin ja.

Hakai:
Saboda haka jini yake ja ne saboda haemoglobin da iron da ke dauke da oxygen a cikin kwayoyin jini.

Aiwatarwa:
Wannan tsarin ne ke ba jikinmu damar kai oxygen ga kowane sashi na jiki domin mu samu karfi da rai.
[Full explanation in Hausa with cultural context]

Limitations

Language: Primarily responds in Hausa; may not be suitable for users requiring English responses
Domain: Specialized in STEM topics; may not perform well on general conversation
Cultural Context: Optimized for Hausa cultural contexts; analogies may not translate to other cultures
Quantization: 4-bit quantization may affect precision in some mathematical calculations
Training Data: Limited to 2,640 STEM examples; may not cover all STEM topics comprehensively

Ethical Considerations

Cultural Sensitivity: Model is designed to respect and promote Hausa cultural values
Educational Access: Aims to democratize STEM education for Hausa speakers
Scientific Accuracy: Maintains rigorous scientific accuracy while making concepts accessible
Language Preservation: Promotes pure Hausa language use in technical domains

Citation

If you use this model in your research, please cite:

bibtex

@misc{amini2024,
  title={AMINI: Culturally-Embedded Hausa STEM Assistant},
  author={Tushe – The Foundry Research Team},
  year={2025},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/AlaminI/AMINI-Culturally-Embedded-STEM-Hausa}}
}

Acknowledgments

Base Model: Meta's Llama-3-8B-Instruct
Training Framework: Unsloth AI for efficient fine-tuning
Dataset: Hausa STEM Reasoning with Cultural Context
Team: Tushe – The Foundry Research Team

License

This model inherits the Llama 3 license from the base model. Please review the Llama 3 License for usage terms.

Contact

For questions, feedback, or collaboration:

HuggingFace: @AlaminI
Dataset: hausa-stem-reasoning-with-cultural-context

Tshe - The Foundry Research (AMINI) - Bridging Scientific Excellence with Cultural Wisdom
Making STEM accessible, relevant, and empowering for Hausa speakers

Tushe/AMINI-Culturally-Embedded-STEM-Hausa