ganda-gemma-1b-GGUF/README.md

---
base_model: google/gemma-3-1b-it
language:
- en
- lg
library_name: transformers
license: gemma
tags:
- luganda
- translation
- conversational
- gemma
- gemma3
- fine-tuned
pipeline_tag: text-generation
---

# 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.

## 📊 Translation Performance

![Translation Performance Comparison](ganda_gemma_ascending_chart.png)

### 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 6.99 BLEU per billion parameters (highest efficiency ratio)  
🚀 **Size Advantage**: Outperforms Gemma 3 4B (4x larger) by 535% on BLEU score  
💎 **Competitive Quality**: Achieves similar performance to GPT-5 Mini with known 1B 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



## 🚀 Quick Start

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained("CraneAILabs/ganda-gemma-1b")
tokenizer = AutoTokenizer.from_pretrained("CraneAILabs/ganda-gemma-1b")

# Translate to Luganda
prompt = "Translate to Luganda: Hello, how are you today?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.3)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```

## 🌍 Language Capabilities

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

## 🎯 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

## 💻 Usage Examples

### Basic Translation

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

model = AutoModelForCausalLM.from_pretrained("CraneAILabs/ganda-gemma-1b")
tokenizer = AutoTokenizer.from_pretrained("CraneAILabs/ganda-gemma-1b")

# English to Luganda translation
prompt = "Translate to Luganda: Welcome to our school"
inputs = tokenizer(prompt, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_length=100,
        temperature=0.3,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

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

### Luganda Conversation

```python
# Direct Luganda conversation
prompt = "Oli otya! Osobola okuntuyamba leero?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.3)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```

### Using the Pipeline

```python
from transformers import pipeline

# Create a text generation pipeline
generator = pipeline(
    "text-generation",
    model="CraneAILabs/ganda-gemma-1b",
    tokenizer="CraneAILabs/ganda-gemma-1b",
    device=0 if torch.cuda.is_available() else -1
)

# Generate Luganda text
result = generator(
    "Translate to Luganda: Welcome to our school",
    max_length=100,
    temperature=0.3,
    do_sample=True
)
print(result[0]['generated_text'])
```

### Ollama Usage

```bash
# Run with different quantizations
ollama run crane-ai-labs/ganda-gemma-1b:q4-k-m    # Recommended balance
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
```

## 🚀 GGUF Advantages

- **Fast CPU Inference**: Optimized for CPU-only environments
- **Memory Efficient**: Multiple quantization levels available
- **Easy Integration**: Works with llama.cpp, Ollama, and other GGUF-compatible tools
- **Cross-Platform**: Supports Windows, macOS, and Linux

## 📦 Available Quantizations

| Quantization | File Size | Memory Usage | Quality | Use Case |
|-------------|-----------|--------------|---------|----------|
| F16         | ~2.0GB    | ~3.5GB      | Best    | GPU inference |
| Q8_0        | ~1.1GB    | ~2.0GB      | High    | High-end CPU |
| Q4_K_M      | ~600MB    | ~1.2GB      | Good    | **Recommended** |
| Q4_K_S      | ~550MB    | ~1.1GB      | Decent  | Low memory |
| Q4_0        | ~550MB    | ~1.0GB      | Basic   | Very low memory |

## 🔗 Related Models

- **Original Model**: [CraneAILabs/ganda-gemma-1b](https://huggingface.co/CraneAILabs/ganda-gemma-1b) - Full precision HuggingFace model
- **Mobile (LiteRT)**: [CraneAILabs/ganda-gemma-1b-litert](https://huggingface.co/CraneAILabs/ganda-gemma-1b-litert) - Optimized for Android/iOS

## 📄 License

This model is released under the [Gemma Terms of Use](https://ai.google.dev/gemma/terms). Please review the terms before use.

---

**Built with ❤️ by Crane AI Labs**

*Ganda Gemma - Your helpful Luganda AI companion, optimized for CPU!*