Multi-lingual LLM Evaluation and Alignment

Evaluated alignment methods for large language models (phi-4, Llama3-8B, and tinyroberta) across 10+ languages, focusing on fairness, robustness, and safety. Employed SFT (Supervised Fine-Tuning) fine-tuning leading to ~15% increase in accuracy across selected languages for each model.

Technologies: Python, PyTorch, HuggingFace Transformers, Multi-lingual NLP, SFT Fine-tuning

Key Achievements:

• Evaluated LLMs across 10+ languages
• Achieved ~15% accuracy improvement through SFT fine-tuning
• Focused on fairness, robustness, and safety metrics