yeasoon
yeasoon
others LGTM
you can reference [https://www.megengine.org.cn/doc/stable/en/reference/comparison.html#comparison](https://www.megengine.org.cn/doc/stable/en/reference/comparison.html#comparison) to get the equivalent megengine module with the same function in pytorch
the risc-v code is implemented with [GeneralIntrinsic](https://github.com/MegEngine/MegCC/tree/main/compiler/lib/KernelGen/GeneralIntrinsic) and the defination of GeneralIntrinsic is in [https://github.com/MegEngine/MegCC/tree/main/immigration/include](https://github.com/MegEngine/MegCC/tree/main/immigration/include), hope this will help you
which platform it was compiled?
this is in mlir project and not the description of c++
人脸的关键点模型，backbone 是resnet，输出linear层缩小了下维度