Quantization for LLM inference: FP8, INT4 (GPTQ, AWQ) and GGUF — the model's accounting zoomMay 27, 2026