You are an engineer on the TensorRT team tasked with developing algorithms for inference on neural networks. A customer has provided a list of networks they are using for inference, and most of them involve a tensor reformat operation. Namely, the reformat operation involves taking in a tensor containing floating-point numbers and outputting a similarly shaped tensor with low-precision integers. You start off with a proof of concept for this functionality in a C++ program.
Task
Reformat tensor with given scale factor
Create a function that takes in an FP32 tensor and floating point scale factor as input. Output an Int8 tensor by converting values using division by the given scale factor.
这道题考察张量数值转换与定点量化的基础实现：给定一个 FP32 tensor 和缩放因子，需要将每个元素按 scale 进行除法后转换为 Int8 tensor，并保持原有形状不变。解题时通常直接遍历张量元素，完成浮点到整型的转换即可，重点在于理解“按比例缩放后再转成低精度整数”的量化思路，以及处理好数据类型转换带来的精度与截断问题。
