为什么生成的耗时非常久，平均一个词要1-2秒，整个句子生成用了十几分钟。

你好，我使用的2张A100*40G，用的是你们14B的大模型执行webdemo.py，交互体验很差，问一个问题要生成答案好久，平均一个回复要十几分钟。请问你们测试也是这样情况吗？

Start by reading and running webdemo.py with the reported 14B model configuration on two A100 40G GPUs. Measure token and full-response latency, then compare the observed behavior with the expected experience described in the issue. Done means the cause of the slow generation is identified and a focused fix or clear performance diagnosis is documented.

Written by the indexing model from the issue text.