Follow Datanami:
August 4, 2021

Spell Operationalizes Advanced AI with a Comprehensive MLOps Platform for Deep Learning

NEW YORK, Aug. 4, 2021 — Spell, a leader in operationalizing AI for natural language processing (NLP), machine vision, and speech recognition, has launched a cloud-agnostic, end-to-end MLOps platform for deep learning. The namesake solution — developed by AI industry veterans — tracks, manages, and automates the entire deep learning workflow, from developing and training, to deploying and optimizing models at scale. The platform dramatically improves efficiency, effectiveness, and compliance for AI projects in large enterprises and AI startups alike.

Current MLOps solutions are designed for traditional machine learning, which uses relatively simple processes and infrastructure, and commodity CPU-based compute resources for building, training, and deploying predictive models. Deep learning can require tracking and managing hundreds of simultaneous experiments with thousands of parameters, and can use vast amounts of more costly GPU-based computation. Without effective MLOps for deep learning (DLOps), most organizations struggle to successfully operationalize new models for NLP, machine vision, voice recognition, and other deep learning applications, or they do so at a very high cost.

Spell addresses these needs by managing and automating the orchestration, documentation, optimization, deployment, and monitoring of deep learning models throughout their entire lifecycle. Ad hoc point tools for those functions work individually and do not enable a comprehensive, integrated view of all the steps in the end-to-end development and deployment workflow. Spell captures and tracks detailed information for every part of the model lifecycle, including the data, parameters, experiments, personnel, compute resources, infrastructure, deployment location, and more. This ensures accountability, collaboration, and productivity, and enables reproducibility, explainability, and governance across the lifespan of the deep learning project.

“We use Spell to orchestrate our production DL payloads so we can focus on creating high-quality models. The flexibility and reliability that Spell provides helps us scale to build hundreds of models every day,” said Zohaib Ahmed, CEO of neural text-to-speech solution provider, Resemble.AI.

Spell also introduces unique cloud infrastructure automation that can reduce compute costs by 66% or more. Spell replaces costly “on-demand” cloud instances by transparently automating the use of far less expensive “spot” instances for continuous training execution. Training and retraining deep learning models using on-demand cloud instances can cost many thousands or even millions of dollars. Spell’s unique virtual on-demand instances can cut those compute costs in half.

Spell is the product of lessons learned by Spell’s founders’ successes in leading the development and use of deep learning management tools and infrastructure at Facebook AI Research (FAIR), Ebay, and machine vision leader, Clarifai. Spell was founded to bring state-of-the-art deep learning MLOps to the masses offering an affordable solution for AI startups, enterprises, and service providers of all sizes, and free access for students and individual practitioners through the Spell Community. Today, Spell’s customers include Akasha, AlphaSense, Cadmium, Condé Nast, Healx, Mulberry, Originate, Quill, RESEMBLE.AI, Whatnot, and Square.

“The rapid growth of the Spell user community is a gratifying validation of our vision for democratizing deep learning through a comprehensive, transparent platform for enabling and accelerating the successful adoption of advanced AI across a broad range of industries and use cases,” said Spell’s CEO and co-founder, Serkan Piantino, adding, “and we are just getting started.”

For more information about the Spell DLOps platform, please visit: https://spell.ml/

About Spell

Founded in 2017 by world-renowned AI experts, Spell operationalizes deep learning at scale with its unique DLOps platform, which is rapidly gaining adoption and recognition for its ability to improve the time-to-value and ROI for computer vision, NLP, and speech recognition, and other advanced AI applications. Spell achieved $15 million in Series A funding led by Two Sigma Ventures and Eclipse Ventures. Learn more: https://spell.ml/.


Source: Spell 

Datanami