关于量化模型推理

非常赞的工作！有两个问题问题请教一下：
1）使用 decoupleQ 量化后的LLaMA模型，推理性能如何？是否有对应一些测试数据？
2）量化后的模型如何部署进行推理呢? README 中，有看到 NVIDIA/TensorRT-LLM#1568，是直接使用 TensorRT-LLM 进行部署吗？然后请问有对应的推理部署脚本吗？

谢谢！

Start with the README references to decoupleQ, LLaMA, and NVIDIA/TensorRT-LLM#1568. Determine whether the repository contains benchmark results or deployment scripts, then document the available inference performance and deployment path; if those artifacts are absent, the scope needs clarification before work begins.

Written by the indexing model from the issue text.