LLM Fine-Tuning & AI Training (BioQwen-NEET)
Fine-tuned a Qwen2.5-3B LLM on a domain-specific NEET Biology corpus using advanced AI training techniques. Conducted supervised fine-tuning on 16K curated samples and implemented reinforcement learning with custom reward functions for correctness and reasoning. Developed structured tool-calling skills within the model using JSON-formatted tags to improve its reasoning evaluation capability. • Managed large-scale, domain-specific text data in accordance with strict curriculum guidelines. • Focused on model alignment and robust performance through supervised and reinforcement learning methods. • Evaluated model performance using task-specific metrics and validation sets. • Engineered custom reward functions to enhance answer quality and reasoning ability.