| --- |
| license: apache-2.0 |
| language: |
| - ta |
| --- |
| |
| # **Debug Divas: Colloquial Tamil Translation Model** |
| 🚀 **Fine-tuned Mistral-7B for English-to-Colloquial Tamil Translation** |
|
|
|  |
| [](https://opensource.org/licenses/MIT) |
|
|
| ## 🌟 **Overview** |
| This model is a **fine-tuned version of Mistral-7B** using **Unsloth's FastLanguageModel**, designed specifically to translate **English text into colloquial Tamil** (spoken Tamil). It is optimized for **real-world Tamil conversations**, making it useful for chatbots, assistants, and translation tools. |
|
|
| --- |
|
|
| ## 📖 **Model Details** |
| - **Base Model**: [Mistral-7B-Instruct](https://huggingface.co/mistral-7b-instruct) |
| - **Fine-Tuned Dataset**: Custom dataset (`debug_divas_dataset.json`) with **English → Colloquial Tamil** translation pairs. |
| - **Training Library**: [Unsloth](https://github.com/unslothai/unsloth) (optimized training for large models) |
| - **Max Sequence Length**: 128 tokens |
| - **Batch Size**: 8 |
| - **Epochs**: 3 |
| - **Optimizer**: AdamW |
|
|
| --- |
|
|
| ## 🔧 **Installation & Setup** |
| To use this model, install the necessary dependencies: |
|
|
| ```bash |
| pip install torch transformers datasets unsloth accelerate |
| ``` |
|
|
| --- |
|
|
| ## 🚀 **Usage** |
| ### **Load Model & Tokenizer** |
| ```python |
| import torch |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
| |
| # Load fine-tuned model from Hugging Face |
| model_name = "your-huggingface-username/debug-divas-tamil-translation" |
| device = "cuda" if torch.cuda.is_available() else "cpu" |
| |
| model = AutoModelForCausalLM.from_pretrained(model_name).to(device) |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| |
| # Translation function |
| def translate_english_to_tamil(input_text): |
| instruction = "Translate the following English sentence to colloquial Tamil" |
| inputs = tokenizer(f"{instruction}: {input_text}", return_tensors="pt").to(device) |
| |
| translated_tokens = model.generate(**inputs, max_length=128) |
| translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True) |
| |
| return translated_text |
| |
| # Example usage |
| input_text = "The pharmacy is near the bus stop." |
| translated_text = translate_english_to_tamil(input_text) |
| print("Colloquial Tamil:", translated_text) |
| ``` |
|
|
| --- |
|
|
| ## 📝 **Example Outputs** |
| | **English** | **Colloquial Tamil Translation** | |
| |------------|---------------------------------| |
| | "How are you?" | "நீங்க எப்படி இருக்கீங்க?" | |
| | "I am going to the market." | "நான் மார்க்கெட்டுக்கு பொறேன்." | |
| | "The pharmacy is near the bus stop." | "மருந்துக் கடை பஸ்ஸ்டாப் அருகே இருக்க." | |
|
|
| --- |
|
|
| ## 📚 **Dataset** |
| The dataset contains **pairs of English sentences** with their **colloquial Tamil translations**. |
| Example format: |
|
|
| ```json |
| [ |
| { |
| "input": "How are you?", |
| "output": "நீங்க எப்படி இருக்கீங்க?" |
| }, |
| { |
| "input": "I am going to the market.", |
| "output": "நான் மார்க்கெட்டுக்கு பொறேன்." |
| } |
| ] |
| ``` |
|
|
| --- |
|
|
| ## 🏗 **Training Details** |
| The model was fine-tuned using **UnslothTrainer** with the following hyperparameters: |
|
|
| - **Batch Size**: 8 |
| - **Epochs**: 3 |
| - **Learning Rate**: 2e-5 |
| - **FP16 Training**: Disabled |
| - **Optimizer**: AdamW |
| - **Dataset Split**: 80% Train, 20% Test |
|
|
|
|
|
|
| --- |
|
|
| ## ⚖ **License & Citation** |
| This model is released under the **MIT License**. If you use it in your work, please cite: |
|
|
| ```bibtex |
| @misc{debugdivas2025, |
| author = {Debug Divas}, |
| title = {Fine-tuned Mistral-7B for Colloquial Tamil Translation}, |
| year = {2025}, |
| publisher = {Hugging Face}, |
| url = {https://huggingface.co/your-huggingface-username/debug-divas-tamil-translation} |
| } |
| ``` |
|
|
| --- |
|
|
| Dataset Linke : https://huggingface.co/datasets/anitha2520/debug_divas45/tree/main |
| anitha2520/debug_divas45 |
|
|
| ## ❤️ **Contributions & Feedback** |
| We welcome feedback and contributions! Feel free to open an issue or contribute to our dataset. |
|
|
| 📧 **Contact:** [xidanitha@gmail.com] |
|
|
| --- |
|
|
|
|
|
|
|
|
