The 4-bit model with superior performance: Quantization‑Aware Healing method compresses and refines large models **Reasoning and explanation** - The Persian phrase “مدل ۴ بیتی با عملکرد برتر” directly translates to “the 4‑bit model with superior performance.” - “روش Quantization‑Aware Healing” is a mixed‑language term where “روش” means “method” and the English technical term “Quantization‑Aware Healing” is kept unchanged. - “مدل‌های بزرگ را فشرده و دقیق می‌کند” means “compresses and refines large models,” where “فشرده” = “compresses” and “دقیق می‌کند” = “makes precise/refines.” - Combining these parts yields the natural English sentence presented above.

Researchers in artificial intelligence have recently introduced the Quantization‑Aware Healing (QAH) method, which intelligently combines the quantization process with weight improvement. This approach allows large models to be compressed to 4‑bit, while reducing prediction error and significantly increasing execution speed.

In comparative experiments, the compressed model with QAH achieved higher scores on standard benchmarks such as GLUE and SuperGLUE compared to the original 32‑bit version. Additionally, memory consumption decreased by up to eightfold and deployment costs on edge servers dropped noticeably. This achievement can pave the way for the development of low‑cost, scalable AI applications.