This week, SiMa.ai dropped the newest Developer Preview release of our Palette ML Developer tools, purpose-built to power the most complex ML challenges with ease. This release represents the next step in fulfilling the technical vision we announced in late January: any model or pipeline, 10x more performant, at the push of a button i.e. “pushbutton.” In April, we established a new standard for performance with our historic MLPerf results (as reported by Forbes: SiMa.ai Sets A New Standard In Embedded Edge Power Efficiency). Now, this preview release delivers on the any and pushbutton promises with a laser focus on the features and capabilities most impactful for modern ML development.
Developers know that before they design anything, they need to prove that the concept itself will actually work. Productization is always preceded by prototypes, most commonly called a Proof of Concept in our industry. These are then used to evaluate their algorithm and capabilities in an edge environment, without dependency on the cloud or any other large hosted platform. This might sound simple, but in reality it presents a major challenge in the world of Machine Learning. ML developers need to be able to port any algorithm pipeline, develop models trained for their own specific use cases and then implement these models in a real-time system where they can interact with real data streams. It’s an involved process, but a necessary one. PoCs often act as both a vision and a roadmap for the final product. At the edge, this eventually has to be in a form factor that can operate with size, weight and power limitations, so even the PoC needs some validation that is possible when developing the actual product.
Time is of the essence for developers. PoCs need to get done, but the faster they’re demonstrated, the faster a developer team can get their vision into production and into the hands of customers. This developer release was built with features and capabilities designed specifically for creating computer vision ML pipelines with lower requirements for code development. This is intended to reduce the friction associated with a PoC, without resorting to hand coding or deeply embedded programming environments. Now, developers can cut PoC efforts down from months to weeks – even days.
How does it work?
The key technology that enables developers to port any model is the SiMa.ai ML compiler, which produces executable code that runs heterogeneously across multiple processors on SiMa.ai’s Machine Learning System on Chip (MLSoC). This, in turn, enables each layer of the model to execute on a data type suitable for the accuracy of that particular layer. This layer partitioning is automatically performed, both in the identification of the layers needed for execution on different processors, and the actual coding and scheduling of these processors to execute seamlessly from the developer’s perspective. This means that the developer is utilizing the compiler in a pushbutton fashion, not only compiling, but also scheduling, deploying and executing the resultant model on the silicon for real time demonstration and analysis. The complexity of this implementation is shielded from the developer, so a data scientist can produce a low-friction PoC even if their team doesn’t have deep expertise in embedded programming.
Why is this so important?
Consumer expectations have evolved, driven by ubiquitous technologies like smart phones and laptops. In the modern B2B environment, innovators want to develop products that demonstrate that same level of capability, usability and refinement. The challenge is that these mass targeted products required thousands of low level developers to support the platform before a single application could be programmed and supported. Any product that cannot sell millions of units is trapped between these reference products, and the reality that this level of development effort is often not feasible. Edge ML products will provide solutions for many unique and different challenges in industrial automation, construction, agriculture, manufacturing, assembly, sorting, inspecting, shipping and handling throughout supply chains, but these products must be designed with a fraction of the resources to reach the full breadth of their potential. To even get prototypes working can take significant resources, before funding for full scale product development is even approved. SiMa.ai’s commitment to support any computer vision application with a low-code environment is key to our vision of scaling ML at the edge, with new and emerging products able to leverage this faster path to a PoC and production.
How do we help developers today?
Developers who understand the difficulty of programming edge silicon devices are already intrigued by what we are delivering. Our Palette platform offers:
- Faster time to value. Understand tools flow, features and capabilities. Build, create and deploy in minutes.
- Versatility. Tackle any model, any computer vision problem imaginable. Auto-partition and compile across MLA and ARM processors.
- Simplicity. Automation is critical to ML development at the edge, eliminating the need for hand coding with push button ease.
- Performance. Exponential performance/Watt gains beat legacy solutions designed for the data center. (See ML Perf win referenced above!).
CV/ML developers can evaluate our Palette software and MLSoC hardware by acquiring a Developer Kit, work with us to develop a PoC based on their requirements or work with their third party ML software solutions provider to port the desired PoC to our silicon platform. To learn more or to purchase our development kit, please click here. We look forward to meeting your ML needs and have continued innovations planned in the near future so stay tuned.