Using embedded simulators in Amazon Braket Hybrid Jobs; AWS a Platinum Sponsor/Exhibitor at IQT Quantum Enterprise in San Diego May 10-12, 2022
(AmazonBlog) Amazon launched a new feature in Amazon Braket Hybrid Jobs on May 5 which allows you to run hybrid workloads with simulators that are embedded with your algorithm code. For instance, one of the simulators available in this new feature is the PennyLane Lightning GPU simulator, accelerated by NVIDIA’s cuQuantum library. IQT-News shares the blog post here.
Amazon will show you how to get started with a “Hello, world” example, and demonstrate some advanced features, such as how to parallelize quantum machine learning workloads across multiple GPU instances for state-of-the-art performance.
Hybrid quantum-classical algorithms, which combine classical and quantum compute resources to iteratively find solutions to computational problems, are a popular approach to near-term quantum computing. In this approach, a parameterized quantum circuit is iteratively adjusted based on intermediate results from a quantum processing unit (QPU) in a classical optimization loop, similar to the training of machine learning models. Researchers in academia and industry are studying the properties of their algorithms in order to answer important questions about the scaling and convergence behavior under different conditions. Since current generation QPUs are limited in their size and coherence, simulators are an important tool for researchers, not only for testing and debugging, but also to study new approaches under ideal conditions, as well as with controlled noise parameters.
With last year’s launch of Hybrid Jobs, we have introduced a framework for the execution of hybrid quantum algorithms on Amazon Braket. As part of your job, you can select any of Braket’s QPUs or on-demand simulator devices, including SV1, DM1, and TN1, to execute the quantum circuits of your hybrid algorithm. With today’s launch, you can now also select embedded simulators for this purpose, where the simulation is part of the algorithm code and executed on the job instance directly. By bringing the simulation closer to your algorithm, you can reduce the latency overhead of communicating with a remote device, which can be a bottleneck for workloads with low qubit numbers (typically up to ~20-25 qubits). Embedded simulators also allow for advanced simulation strategies, such as adjoint differentiation for more efficient gradient computations. Lastly, through the Bring Your Own Container (BYOC) feature, you can experiment with and develop other open source simulation libraries.
Click thru to the blog to learn how you can get started with embedded simulators in Braket Hybrid Jobs!
Sandra K. Helsel, Ph.D. has been researching and reporting on frontier technologies since 1990. She has her Ph.D. from the University of Arizona.