Connecting Frameworks

While a Deep Learning framework is ultimately meant for end use by data scientists, or for deployment in cloud container environments, nGraph Core ops and the nGraph C++ Library are designed for framework builders themselves. We invite anyone working on new and novel frameworks or neural network designs to explore our highly-modular stack of components that can be implemented or integrated in virtually limitless ways.

Please read the articles in this section if you are considering incorporating components from the nGraph Compiler stack in your framework or neural network design. Articles here are also useful if you are working on something built-from-scratch, or on an existing framework that is less widely-supported than the popular frameworks like TensorFlow and PyTorch.

Understanding users of frameworks

A data scientist or ML engineer may not initially know which framework is the “best” framework to use to start working on his or her problem set. While there are several to choose from, it can be daunting and time consuming to scope the wide array of features and customization options offered by some of the more popular frameworks:

  1. First find a tested and working DL model that does something similar to what the data scientist or ML engineer wants to do. To assist with this stage, we’ve already provided organized tables of Validated Models and Workloads examples.

  2. Next, replicate that result using well-known datasets to confirm that the model does indeed work. To assist with this stage, we’ve released several pip installation options that can be used to test basic examples.

  3. Finally, modify some aspect: add new datasets, or adjust an algorithm’s parameters to hone in on specifics that can better train, forecast, or predict scenarios modeling the real-world problem. This is also the stage where it makes sense to tune the workload to extract best performance.

    Important

    nGraph does not provide an interface for “users” of frameworks (for example, we cannot dictate or control how Tensorflow* or MXNet* presents interfaces to users). Please keep in mind that designing and documenting the User Interface is entirely in the realm of the framework owner or developer and beyond the scope of the nGraph Compiler stack. However, any framework can be designed to make direct use of nGraph Compiler stack-based features and then expose an accompanying UI, output message, or other detail to a user.

Clearly, one challenge of the framework developer is to differentiate from the pack by providing a means for the data scientist to obtain reproducible results. The other challenge is to provide sufficient documentation, or to provide sufficient hints for how to do any “fine-tuning” for specific use cases. With the nGraph Compiler stack powering your framework, it becomes much easier to help your users get reproducible results with nothing more complex than the CPU that powers their operating system.

In general, the larger and more complex a framework is, the harder it becomes to navigate and extract the best performance; configuration options that are enabled by “default” from the framework side can sometimes slow down compilation without the developer being any the wiser. Sometimes only a few small adjustments can increase performance. Likewise, a minimalistic framework that is designed around one specific kind of model can sometimes offer significant performance-improvement opportunities by lowering overhead.

See Configurations available to any framework to get started.