---
language:
- zu
- en
base_model: google/gemma-3-4b-it
tags:
- machine-translation
- isizulu
- african-languages
- gemma
- peft
- lora
datasets:
- lelapa/Inkuba-instruct
license: gemma
widget:
- text: "Translate this from isiZulu to English: Sawubona, unjani?"
  example_title: "isiZulu to English"
---

# Simple isiZulu→English Translation Model

Simple and focused isiZulu to English translation model using consistent prompting.

## Model Details

- **Base Model**: google/gemma-3-4b-it
- **Task**: isiZulu → English Translation
- **Training Examples**: 50000
- **Prompt Format**: "Translate this from isiZulu to English: [text]"

## Usage

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

# Load model
base_model = AutoModelForCausalLM.from_pretrained("google/gemma-3-4b-it")
tokenizer = AutoTokenizer.from_pretrained("google/gemma-3-4b-it")
model = PeftModel.from_pretrained(base_model, "Dineochiloane/gemma-3-4b-isizulu-simple")

# Translate
messages = [{"role": "user", "content": "Translate this from isiZulu to English: Ngiyabonga kakhulu"}]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt")
outputs = model.generate(input_ids, max_new_tokens=50, repetition_penalty=1.2)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```

## Training Details

- **LoRA Rank**: 32
- **LoRA Alpha**: 32
- **Learning Rate**: 5e-05
- **Epochs**: 3
- **Consistent Prompting**: All training uses same format as evaluation