Prompt Engineer/Fine-tuner - UIT Data Science Challenge 2025
TextPrompt Response Writing SFT
I fine-tuned Qwen3-4B-Instruct-2507 for hallucination classification in Vietnamese LLM outputs. The process involved preparing high-quality instruction-based SFT prompts with three hallucination classes: no, intrinsic, extrinsic. Evaluation and quality control led to a top 7 ranking in a national data science competition. • Designed and annotated instruction-based classification prompts
• Performed fine-tuning leveraging RLHF guidelines
• Validated annotation quality via F1 metric (0.841)
• Implemented structured evaluation using golden test cases
I fine-tuned Qwen3-4B-Instruct-2507 for hallucination classification in Vietnamese LLM outputs. The process involved preparing high-quality instruction-based SFT prompts with three hallucination classes: no, intrinsic, extrinsic. Evaluation and quality control led to a top 7 ranking in a national data science competition. • Designed and annotated instruction-based classification prompts • Performed fine-tuning leveraging RLHF guidelines • Validated annotation quality via F1 metric (0.841) • Implemented structured evaluation using golden test cases
2025 - 2025