Abstract
In this digital era of Artificial Intelligence, PyTorch is considered as one of the highly used and constantly evolving open-source machine learning libraries in building efficient and high-performance models. However, PyTorch being an evolving framework, the developer community has identified several areas that can be enhanced further. One such area we want to improvise is the Numerical Gradient Checking functionality which plays a prominent role in identifying the correctness of the model is limited only to the data types of float and complex float and cannot apply to the Tensor object, where Tensor is an object which is mainly used to store n-dimensional arrays. As part of this Master Project, I want to enhance the Automatic Differentiation Package "Torch.AutoGrad" of the PyTorch library by expanding the numerical gradient functionality to Tensor object and making it easier for the users to test the correctness of their model.