Deci Launches Version 2.0 of Its Deep Learning Platform to Ease AI Application Development

Deci, a deep learning company that uses AI to build AI, has released Version 2.0 of its deep learning development platform, making it easier than ever for AI developers to build, optimize, and deploy computer vision models with exceptional accuracy and runtime performance on any hardware and environment, including cloud, edge, and mobile.

In a media advisory, Deci CEO and co-founder Yonatan Geifman noted that by leveraging this platform, resultant AI models may be more quickly built to operate on any hardware and environment, including cloud, edge, and mobile accuracy and excellent runtime performance. The DeciNet templates included in the v2.0 platform have eliminated most of the grunge work.

Developing production-ready deep learning models for deployment is difficult for AI engineers. These issues are primarily due to the AI efficiency gap that the industry is experiencing. Algorithms are becoming more powerful and complicated, but processing capacity is not keeping up. This divide also generates financial constraints by making the development and processing of deep learning more difficult and costly.

AI developers may increase inference performance and efficiency using Deci, according to the business, enabling successful deployments on resource-constrained edge devices, maximizing hardware utilization, and lowering training and inference costs, according to Geifman. He claims that the whole development cycle is reduced, which saves money upfront, and that the ambiguity of how the model would run on inference hardware is eliminated.

While Neural Architecture Search (NAS) has been proposed as a possible method for automating the construction of improved artificial neural networks that surpass manually-designed architectures, the resource requirements to run such technology are prohibitive. Yet, only Google, Microsoft, and academic institutions have effectively adopted NAS, demonstrating its impracticality for the great majority of developers.

To address this issue, Deci’s platform, which is driven by their unique NAS engine dubbed AutoNAC (Automated Neural Architecture Construction), allows AI developers to create efficient computer vision models automatically and economically for any given inference hardware, speed, size, and targets. Deci models outperform other known state-of-the-art (SOTA) designs by three to ten.

Developers may utilize the AutoNAC engine to construct more unique architectures adapted for their individual use-cases or start their projects using pre-trained and optimized models (DeciNets) provided by the AutoNAC engine for a wide range of hardware and computer vision activities. 

In addition, the platform provides teams with a variety of tools for developing deep learning-based applications, such as a hardware-aware model zoo that allows them to quickly select and benchmark models and hardware, SuperGradients – an open-source training library with proven recipes for faster training, automated runtime optimizations, model packaging, and more.

AI developers may increase inference performance and efficiency by adopting Deci’s platform to allow deployment on resource-constrained edge devices, maximize hardware usage, and lower training and inference costs. The whole development cycle is reduced, and there is no longer ambiguity about how the model will run on the inference hardware.

Developers may quickly assess the inference time of pre-trained and optimized models on various hardware, including edge devices, using Deci’s hardware-aware model zoo. Eliminate the requirement to manually install and test numerous combinations of models and hardware to simplify the hardware and model selection process.

Deci’s platform is divided into three sections:

  • Data scientists and machine learning developers that want to identify the best models, simplify hardware assessment, and improve runtime performance can use the free Community Tier.
  • Professional Tier: For deep learning teams who want to reach production-grade inference performance rapidly and reduce development time.
  • Deep learning professionals aiming to accomplish particular performance targets for highly specialized use cases should choose the Enterprise Tier.


With Deci’s AutoNAC engine, AI developers can automatically find correct and efficient architectures matched with the application, hardware, and performance requirements. They may use Deci’s PyTorch-based open-source training library, SuperGradients, to use proven hyperparameter formulations. It can assemble and quantify models automatically and analyze various production settings. With Deci’s python-based inference engine, developers can deploy their deep learning workloads in any environment.



Prathamesh Ingle is a Consulting Content Writer at MarktechPost. He is a Mechanical Engineer and working as a Data Analyst. He is also an AI practitioner and certified Data Scientist with interest in applications of AI. He is enthusiastic about exploring new technologies and advancements with their real life applications