可以解释preset_kernels目录下bitnet-lut-kernels-tl2.h文件中查表操作的实现吗？希望能从代码层面解释：1. 如何预处理生成一张表 2. 查表操作的实现

[dujifeng]

[sd983527]

Please refer to this paper which firstly introduced LUT method for low bits model inference. https://arxiv.org/pdf/2407.00088