Qwen2.5-1M Technical Report

Details on training/inference optimizations for million-token contexts (Jan 2025)