Bridging the gap between complex scientific research and the curious minds eager to explore it.

Computer Science, Distributed, Parallel, and Cluster Computing

Reconfiguring CPU-based DNN Serving for Low Latency

Reconfiguring CPU-based DNN Serving for Low Latency

In this article, we present Tensor Comprehensions (TC), a novel framework for high-performance machine learning (ML) abstractions. By leveraging the power of tensor comprehensions, we aim to make ML more accessible and efficient, without sacrificing accuracy. TC is designed to be framework-agnostic, allowing developers to use any ML framework they prefer while still reaping the benefits of TC’s performance enhancements.

Tensor Comprehensions: What are they?

A tensor comprehension is a shorthand for complex mathematical operations on tensors. In essence, it allows developers to define a set of rules that manipulate tensors in a particular way. TC takes this concept further by providing a unified framework for defining and applying tensor comprehensions across different ML frameworks. This means that developers can write tensor comprehensions once and use them seamlessly with various ML frameworks, saving time and effort.

Worker Management: The Key to Efficient ML Computations

One of the primary challenges in scaling ML computations is managing the workers responsible for performing these computations. TC addresses this challenge by introducing a novel worker management layer that efficiently schedules and executes tensor comprehensions across multiple workers. This layer is designed to work seamlessly with any ML framework, ensuring that developers can focus on writing their models without worrying about the underlying infrastructure.

OpenBLAS: The Optimized BLAS Library

Optimizing the Basic Linear Algebraic Subprograms (BLAS) library is crucial for high-performance ML computations. TC leverages OpenBLAS, an open-source optimized BLAS library, to further enhance the performance of TC’s tensor comprehensions. By using OpenBLAS, TC can achieve even faster performance than traditional BLAS libraries, making it an ideal choice for ML applications.
Conclusion: Democratizing High-Performance ML Computations

In conclusion, Tensor Comprehensions offer a powerful solution for democratizing high-performance machine learning computations. By leveraging the power of tensor comprehensions and efficient worker management, TC makes ML more accessible and efficient without sacrificing accuracy. With its support for various ML frameworks and OpenBLAS optimization, TC is poised to revolutionize the field of ML research and development. Whether you’re a seasoned ML expert or just starting out, TC offers an exciting opportunity to enhance your ML capabilities and take your work to the next level.