Tencent Hunyuan compresses its 1.8B translation model for edge devices using 2-bit and 1.25-bit quantization, reducing size to ~574MB and 440MB with near-lossless quality. Deployed on Bilibili for real-time live stream comment translation, the solution uses 500-700MB memory, processes comments in 500-800ms, and operates locally to lower costs and enhance privacy. This marks a breakthrough from lab demo to high-concurrency commercial application across mobile and PC platforms.