Technology Friday: Bitfusion Flex Brings Lifecycle Management to Artificial Intelligence
On today’s technology Friday I would like to cover one of the most exciting platforms in the deep learning space: Bitfusion flex. In the past ( more like every week ;) ) I’ve written about the need for platforms that simplify the management, testing and training of deep learning applications. Bitfusion is trying to tackle this challenge head on with its Flex platform. The company recently received a strong validation by raising $5 million from investors such as Sierra Ventures, Vanedge Capital, Data Collective and several others.
One of the main challenges in the deep learning market is the lack of balance between application development frameworks and the corresponding application management tools. While we have witnessed an explosion on the number and variety of frameworks for building neural network applications such as TensorFlow, Theano, Torch, Caffe, MxNet, Bonsai, PaddlePaddle and many others, the tools and frameworks of managing and monitoring deep learning applications haven’t evolved at the same pace.
Bitfusion Flex provides an elastic, GPU-optimized infrastructure for the execution of AI programs written in frameworks such as TensorFlow, Torch, Caffe, MxNet and others. Flex also includes tools of managing, training and monitoring deep learning applications.
The Bitfusion Manager is the main interface to interact with the Flex platform. The tool provides a web portal for managing deep learning projects and datasets as well as complementary assets. Projects are a core concept of the Bitfusion platform that encapsulates the code and data associated with a deep learning application. Developers can create and execute projects directly from the Bitfusion CLI which provides a command-line interface that automates many aspects of the lifecycle of deep learning applications. As part of the Project configuration, developers can specify the runtime environment (TensorFlow, Torch, Caffe…) to execute the application as well as the expected GPU topology. Bitfusion typically encapsulates projects as containers and attaches GPU-CPU resources to them.
As mentioned before, one of the greatest benefits of Bitfusion Flex are its multi-framework capabilities. In order to streamline the implementation of deep learning applications ,Flex introduces the notion of Workspaces, an interactive environment based on Jupyter . With Workspaces, Bitfusion Flex developers have access to the core Jupyter artifacts such as code notebooks, file browser and the terminal web shell. The Bitfusion CLI is also included as part of the Workspace environment so that developers can manage different aspects of their applications without having to switch tools.
another innovative capability of Bitfusion Flex is the fact that it combines data and code within the same environment. Developers can use the Bitfusion CLI or Manager to upload datasets that are used to train, test and optimize deep learning applications.
Who is the Competition?
Bitfusion Flex is a unique offering in the deep learning market and, as a result, it is still not experiencing a lot of competition. Other multi-runtime deep learning platform such as Floyd or the Alibaba Cloud Machine Learning service can be considered competitors. Also, if we are solely talking about TensorFlow applications, Google Cloud ML provides a comparable feature set to Flex.