Qwen3-1.7B-Tamil-16bit-Instruct

Model Description

This is a fine-tuned version of Qwen3-1.7B specifically optimized for Tamil language tasks. The model has been trained to understand and generate Tamil text across various domains including coding, entertainment, question-answering, reasoning, literature, ethics, and translation.

• Developed by: sabaridsnfuji
• Model type: Causal Language Model
• Language: Tamil
• License: Apache 2.0
• Base model: Qwen3-1.7B
• Parameter count: 1.7B
• Precision: 16-bit

Training Details

This Qwen3 model was trained 2x faster with Unsloth and Hugging Face's TRL library.

Training Dataset

• Dataset: abhinand/tamil-alpaca-orca
• Description: A comprehensive Tamil instruction-following dataset based on Alpaca and Orca methodologies

Evaluation

Evaluation Dataset

• Dataset: abhinand/tamil-llama-eval
• Evaluation Date: 2025-07-20
• Total Samples: 466

Overall Performance Metrics

Metric	Score	Standard Deviation
Overall Quality	0.704	0.032
Fluency	0.914	0.023
Relevance	0.565	0.078
Coherence	0.371	0.061
Completeness	0.750	0.039
Safety Score	0.984	0.009
Hallucination Risk	0.002	0.004
Perplexity	174.942	904.409

Category-wise Performance

Category	Samples	Overall Quality	Fluency	Relevance	Safety
Entertainment	50	0.749	0.911	0.711	0.974
Reasoning	50	0.740	0.920	0.574	0.968
Open QA	50	0.722	0.933	0.656	0.984
Literature	50	0.718	0.921	0.597	0.992
QA	50	0.711	0.909	0.556	0.980
Ethics	50	0.700	0.921	0.562	0.992
Generation	50	0.695	0.926	0.524	0.996
Unknown	16	0.690	0.894	0.529	1.000
Translation	50	0.664	0.937	0.462	0.976
Coding	50	0.642	0.855	0.451	0.988

Key Strengths

✅ High Overall Quality: Achieves 0.704 overall quality score, meeting recommended standards
✅ Excellent Fluency: Strong fluency score of 0.914 across all categories
✅ Superior Safety: Very high safety score of 0.984 with minimal hallucination risk (0.002)
✅ Best Performance: Excels in entertainment content generation (0.749 quality score)
✅ Low Hallucination Risk: Extremely low hallucination risk of 0.002

Areas for Improvement

📊 Coherence: Moderate coherence score (0.371) could benefit from improvement
📊 Coding Tasks: Lower performance in coding category (0.642) - area for future enhancement

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct")
tokenizer = AutoTokenizer.from_pretrained("sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct")

# Example usage
prompt = "உங்கள் கேள்வி இங்கே:"  # Your question here:
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Intended Use

This model is designed for:

• Tamil text generation and completion
• Question-answering in Tamil
• Entertainment content creation
• Literature and creative writing
• General conversation in Tamil
• Translation tasks (with noted limitations)

Limitations

• Coding performance is below optimal levels
• Coherence scores indicate room for improvement in maintaining logical flow
• Translation tasks show lower relevance scores
• Performance may vary significantly across different domains

Ethical Considerations

The model maintains high safety standards (0.984) and extremely low hallucination risk (0.002), making it suitable for responsible AI applications. However, users should always review outputs for accuracy, especially for critical applications.

Citation

If you use this model, please cite:

@misc{qwen3-tamil-instruct,
  title={Qwen3-1.7B-Tamil-16bit-Instruct},
  author={Sabari Nathan},
  year={2025},
  url={https://huggingface.co/sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct}
}