A user tested the 4-bit version of GLM-5.2 (GLM-5.2-UD-Q4_K_XL) on a server equipped with an Epyc Rome 7452 processor and 512GB of RAM. The model was evaluated using a complex coding prompt requiring the creation of a self-contained 3D arena game in HTML, CSS, and JavaScript.

  • Hardware: AMD Epyc Rome 7452 CPU with 512GB RAM.
  • Model: GLM-5.2-UD-Q4_K_XL (4-bit quantization).
  • Task: Generate a single .html file for a 3D game using Three.js, including player movement, collectibles, enemies, and HUD.
  • Output: The model generated 15,510 tokens over a period of 2 hours and 29 minutes.
  • Quality: The author noted surprise at the high quality of the generated code.

The test demonstrates that large language models can perform complex reasoning and code generation tasks on CPU-only hardware given sufficient memory.