Optimize memory usage for T4 GPU training

Fix: Add required YAML frontmatter to README for HF Spaces

Add Qwen2.5-0.5B fine-tuning on Codeforces CoTs

initial commit