GLM 5.2 is 753B params and 1.51TB at full precision. Run it locally: the live Unsloth quant ladder, every GPU and RAM path, and the quant to actually target.