An End-to-End OCR pipeline for the AMD AI Challenge using Qwen-VL, PyTorch, and ROCm. Features strict GPU VRAM optimization through static execution graphs and Docker containerization.