Ganda Gemma 1B

A fine-tuned Gemma 3 1B instruction model specialized for English-to-Luganda translation and Luganda conversational AI. The model accepts input in both English and Luganda but outputs responses exclusively in Luganda.

🚀 Quick Start

# Run the recommended Q4_K_M quantization
ollama run crane-ai-labs/ganda-gemma-1b:q4-k-m

# Try different quantizations based on your needs
ollama run crane-ai-labs/ganda-gemma-1b:q8-0    # Higher quality
ollama run crane-ai-labs/ganda-gemma-1b:q4-k-s  # Smaller size
ollama run crane-ai-labs/ganda-gemma-1b:f16     # Original quality

🌍 Language Capabilities

• Input Languages: English + Luganda
• Output Language: Luganda only
• Primary Focus: English-to-Luganda translation and Luganda conversation

📊 Performance Metrics

Model Comparison

Model	Parameters	BLEU	chrF++	Efficiency*
Gemma 3 4B	4B	1.1	20.05	0.28
Gemma 3 27B	27B	3.65	31.37	0.14
GPT-5 Mini	N/A	5.14	36.55	N/A
Ganda Gemma 1B	1B	6.99	40.32	6.99
Gemini 2.0 Flash	Large	7.94	43.38	N/A

*Efficiency = BLEU Score / Parameters (in billions)

Key Performance Insights

🏆 Efficiency Leader: Achieves the highest BLEU-to-parameter ratio (6.99 BLEU per billion parameters)
🚀 Size Advantage: Outperforms Gemma 3 4B (4x larger) by 535% on BLEU score
💎 Competitive Quality: Outperforms GPT-5 Mini by 36% on BLEU score with known parameter count
⚡ Practical Deployment: Runs efficiently on consumer hardware while maintaining quality

Evaluation Details

• Dataset: FLORES-200 English→Luganda (1,012 translation pairs)
• Metrics: BLEU (bilingual evaluation understudy) and chrF++ (character F-score)
• Evaluation: Zero-shot translation performance

📊 Available Quantizations

Quantization	Size	Quality	Use Case
f16	~1.9GB	Highest	Maximum quality inference
f32	~3.8GB	Highest	Research & benchmarking
q8-0	~1.0GB	Very High	Production with ample resources
q5-k-m	~812MB	High	Balanced quality/size
q4-k-m	~769MB	Good	Recommended for most users
q4-k-s	~745MB	Good	Resource-constrained environments
q3-k-m	~689MB	Fair	Mobile/edge deployment
q2-k	~658MB	Lower	Minimal resource usage

🤖 Model Details

• Base Model: Gemma 3 1B Instruction Tuned
• Specialization: English-to-Luganda translation and Luganda conversation
• Context Length: 32K tokens
  
• Architecture: Transformer with sliding window attention
• Input Languages: English + Luganda
• Output Language: Luganda only

⚙️ Generation Parameters

The model is optimized with the following parameters:

temperature: 0.3      # Focused, coherent responses
top_p: 0.95          # Nucleus sampling
top_k: 64            # Top-k sampling  
max_tokens: 128      # Response length limit
repeat_penalty: 1.1  # Reduces repetition

💻 Usage Examples

Basic Translation

ollama run crane-ai-labs/ganda-gemma-1b:q4-k-m
>>> Translate this to Luganda: "Hello, how are you today?"

Luganda Conversation

ollama run crane-ai-labs/ganda-gemma-1b:q4-k-m
>>> Oli otya! 

With Custom System Prompt

import ollama

response = ollama.chat(
    model='crane-ai-labs/ganda-gemma-1b:q4-k-m',
    messages=[
        {
            'role': 'system',
            'content': 'You are a helpful Luganda conversation assistant.'
        },
        {
            'role': 'user', 
            'content': 'Can you help me learn basic Luganda greetings?'
        }
    ]
)

API Usage

curl http://localhost:11434/api/chat -d '{
  "model": "crane-ai-labs/ganda-gemma-1b:q4-k-m",
  "messages": [
    {
      "role": "user",
      "content": "Translate: Good morning, how did you sleep?"
    }
  ]
}'

🎯 Capabilities

• Translation: English-to-Luganda translation
• Conversational AI: Natural dialogue in Luganda
• Summarization: Text summarization in Luganda
• Writing: Creative and informational writing in Luganda
• Question Answering: General knowledge responses in Luganda

🔧 Technical Specifications

• Model Family: Gemma 3
• Parameters: 1 billion
• Precision: Multiple quantization levels available
• Context Window: 4,096 tokens
• Architecture: Transformer with optimized attention patterns
• Tokenizer: SentencePiece with 262K vocabulary

🏆 Performance

Ganda Gemma 1B delivers: - ✅ High-quality English-to-Luganda translation - ✅ Natural Luganda conversation - ✅ Effective text summarization - ✅ Fast inference on consumer hardware - ✅ Efficient memory usage

🔄 Chat Template

The model uses the official Gemma 3 chat template:

<start_of_turn>user
Your message here<end_of_turn>
<start_of_turn>model
Model response<end_of_turn>

🛠️ Installation & Setup

1. Install Ollama (if not already installed):

   curl -fsSL https://ollama.com/install.sh | sh
   

2. Pull the model:

   ollama pull crane-ai-labs/ganda-gemma-1b:q4-k-m
   

3. Start chatting:

   ollama run crane-ai-labs/ganda-gemma-1b:q4-k-m
   

📚 Integration Examples

Python with ollama-python

import ollama

client = ollama.Client()

response = client.chat(
    model='crane-ai-labs/ganda-gemma-1b:q4-k-m',
    messages=[
        {'role': 'user', 'content': 'Translate to Luganda: Hello!'}
    ]
)

print(response['message']['content'])

JavaScript with ollama-js

import { Ollama } from 'ollama'

const ollama = new Ollama({ host: 'http://127.0.0.1:11434' })

const response = await ollama.chat({
  model: 'crane-ai-labs/ganda-gemma-1b:q4-k-m',
  messages: [{ role: 'user', content: 'Translate to Luganda: Good morning!' }],
})

console.log(response.message.content)

🎨 Use Cases

• Language Learning: English-Luganda translation practice
• Cultural Preservation: Luganda content creation and documentation
• Educational Tool: Learning assistant in Luganda
• Content Localization: Translating content to Luganda
• Conversational Practice: Luganda dialogue practice
• Text Summarization: Summarizing content in Luganda

⚠️ Limitations

• Language Output: Responds only in Luganda
• Factual Knowledge: General knowledge only, not trained on specific factual datasets
• No Coding/Math: Not designed for programming or mathematical tasks
• Knowledge Cutoff: Training data has a knowledge cutoff date
• Context Length: Limited to 4,096 tokens for optimal performance
• Specialized Domains: May require domain-specific fine-tuning for technical fields

📄 License

This model is released under the Gemma Terms of Use. Please review the terms before use.

🤝 Contributing

Found an issue or want to improve the model? We welcome: - Bug reports and feedback - Performance evaluations and benchmarks - Use case examples and integration guides - Documentation improvements

🔗 Links

• Model Page: https://ollama.com/crane-ai-labs/ganda-gemma-1b
• Ollama: https://ollama.com
• Gemma Family: https://ai.google.dev/gemma
• Base Model: google/gemma-3-1b-it

🙏 Acknowledgments

• Google DeepMind: For the Gemma 3 base model, support and guidance.
• Community: For Luganda language resources and datasets
• Gilbert Korir (Msingi AI, Nairobi, Kenya)
• Alfred Malengo Kondoro (Hanyang University, Seoul, South Korea)

Citation

If you use these models in your research or applications, please cite:

@misc{crane_ai_labs_2025,
    author    = {Bakunga Bronson and Kato Steven Mubiru and Lwanga Caleb and Gimei Alex and Kavuma Lameck and Roland Ganafa and Sibomana Glorry and Atuhaire Collins and JohnRoy Nangeso and Tukamushaba Catherine},
    title     = {Ganda Gemma: A Fine-tuned Gemma 3 1B Model for Luganda conversational AI},
    year      = {2025},
    url       = {https://huggingface.co/CraneAILabs/ganda-gemma-1b},
    organization = {Crane AI Labs}
}

---

Built with ❤️ by Crane AI Labs