Chat with us, powered by LiveChat

Product

OnSpecta’s DLS Inference Engine

OnSpecta helps customers achieve superior performance acceleration for AI workloads by deploying its DLS Inference Engine across their existing compute environment.

DLS is a seamless binary drop-in library to any AI framework that accelerates inference without any accuracy loss, conversions, or model retraining. Its architecture is diagrammed in the figure below. The main components of DLS are as follows:

The Framework Integration Layer provides seamless integration to the developer frameworks. We work with the trained networks “as is”. No conversions or approximations are required.
Our Network Optimization Layer implements techniques such as structural network enhancements, changes to the processing order for efficiency, and data flow optimizations, without any accuracy degradation.
The Hardware Acceleration Layer includes a “just-in-time”, platform-independent optimization compiler that talks to a small number of Microkernels which are small bits of hardware-specific code. This approach allows DLS to deliver high-performance and support multiple hardware platforms.
OnSpecta DLS IE Product Breakdown

All model types are supported

Key benefits

Frameworks supported

Tensor Flow
PyTorch
ONYX
Tensor Flow Lite

Hardware platforms supported

ARM
Ampere
Intel
Nvidia

Hardware platforms supported

Tensor Flow
PyTorch
ONYX
Tensor Flow Lite

Frameworks supported

ARM
Ampere
Intel
Nvidia