Deep Learning techniques are applied in software systems rapidly. Therefore, it becomes necessary to specify Deep Learning (DL) application programming interfaces (APIs) for desired output. Unlike traditional software, DL-specific development occupies bugs that exhibit not only crashes but also yield low accuracy and high training time issues. Inspired by the design-by-contract (DbC) methodology, this work proposes a preemptive measure against such bugs, we call it DL Contract. DL Contract aims to document properties of DL libraries and provide developers with a mechanism to prevent low accuracy and high training time-related bugs during development. One of the main challenges towards DL Contract is to specify properties of the training process, which is inaccessible at the functional interface of the DL libraries. Thus, we introduce the notion of ML variable that allows developers to specify the properties of model architecture, data, and training behavior. To evaluate the utility of DL Contract, we intend to utilize benchmarks from prior works on DL bug detection and repair after implementing DL Contract for Python-based DL libraries.