Visit Lightning AI

Tag: scale

Helping Consultants Apply Modern Methods To Industrial Problems

The Company

Willows.ai is a Montreal-based team that comprises machine learning, AI, and software experts who develop and deliver full-stack AI solutions for the manufacturing industry. Manufacturing has become a proving ground for leveraging the benefits of machine learning, where companies use technology to optimize visual systems to monitor and increase safety, track progress and provide real-time status updates, and reduce common process-driven waste. With expertise in computer vision and building robust, explainable machine learning solutions, Willows.ai helps manufacturers apply the latest research to their business problems. 

 

The Problem

Prior to discovering Grid, the Willows.ai team managed a complex system to compare and analyze the results of their machine learning operations. This complexity often led to duplication of work and ballooning computational costs. 

Willows knew that if they wanted to scale, they needed to avoid wasting precious time worrying about cloud infrastructure. They wanted a platform that freed up resource time and delivered results faster for both them and their clients. 

 

Solution 

The Willows team discovered Grid.ai through the open source PyTorch Lightning community. The Grid platform, developed by the same team behind PyTorch Lightning, provided Willows with the capabilities they needed to manage scalable ML workflows. “We were able to solve our need to deliver more value by providing massive scale training to our clients,” explained Dr. Andrew Marble. It is extremely valuable for us to manage and see all the experiments that we create within a single dashboard view.” The Grid platform provides a unified view for managing run metrics, logs, and artifacts.

 

grid platform

 

What Willows loved most about Grid is its Hyperparameter Optimization functionality that enabled them to scale variations of their models in parallel without needing to change their code or take on external dependencies. 

 

“Being able to parameterize and add parameters to your model is where the real value [for Willows] is at. Grid’s Hyperparameter Sweeps are the most interesting feature to me”, Dr. Andrew Marble, Principal, Willows.ai.

 

Grid’s ability to manage interruptible compute such as spot instances through the click of a button reduced training costs for Willow.ai. Receiving real-time estimates before, during, and after, running experiments with Grid provided Willows with the transparency it needed to manage costs.

willows runs

Getting Started With Grid.ai

Interested in learning more about how Grid can help you manage deep learning model development for your next project? Get started with Grid’s free community tier account (and get $25 in free credits!) by clicking here. Also, explore our documentation and join the Slack community to learn more about what the Grid platform can do for you.

 

Get to know more about Willows.ai, and connect with Dr. Andrew Marble via email at andrew@willows.ai

Managing Machine Learning With Limited Resources

Machine Learning Made Easier With Grid.ai

In the race to implement machine learning, businesses of all sizes look for more accurate predictions that will allow them to stay ahead and differentiate themselves from the competition. One of the main roadblocks in getting there is the availability and affordability of machine learning resources. In this post, we speak with Felix Dittrich, the lead Machine Learning Developer at Memoresa to discuss how Grid.ai solved some of their ML challenges.

 

Discovering Grid

Based in Leipzig, Germany, Memoresa is an online platform and mobile app for easy estate planning, secure emergency provision, and digital organization with a system. Memoresa prides itself on helping estates organize their life documents to allow peace of mind for the organizing and planning of life.

Before finding Grid, Felix and his team felt challenged by all the steps related to training machine learning management, mainly due to their limited number of developers, capabilities, and resources brought on by a small team.

Memoresa captures image documents in their application; the ML engineer uses Grid to train their model to capture metadata for each user and auto-complete forms that are required to simplify the onboarding process. “We used Grid to train a custom named entity recognition model using Transformers and PyTorch Lightning as well as Onnx for quantization,” mentioned Felix. 

While using the open-source PyTorch Lightning project to eliminate boilerplate in his code, Felix discovered Grid.ai. Founded by the creators of PyTorch Lightning, Grid is a platform designed to develop and train deep learning models at scale. The team at Memoresa needed an easy way to coordinate all the steps in training and managing machine learning models. Grid made it easy for Felix’s team to address this, as well as version control all their training data and model artifacts out of the box.

 

Memoresa 1

Exceeded Expectations!

Implementing Grid exceeded Felix and his team’s expectations. “All the steps related to end-to-end model training can be managed in one place. When my code runs in Sessions, I can scale it with Runs without any code changes.”

For start-ups like Memoresa, this means that it takes less time to go from an innovative prototype to a top-performing model. Felix remarked that he likes “how easily Grid Sessions enables me to prototype and debug my models and how I can scale my Session code with different hyperparameters configurations with Runs without any code modifications.”

“I love that Grid supports automatic versioning of datastores and model artifacts. ​Grid makes it simple to share datasets and model assets and code. As our company scales, it’s straightforward to introduce new teammates for collaboration.” This means that start-ups such as Memoresa can easily share the outcomes of their machine learning experiments reducing the time from expiration to business value.

Memoresa 2

 

Memoresa 3

Getting Started With Grid.ai

Interested in learning more about how Grid can help you manage deep learning model development for your next project? Get started with Grid’s free community tier account (and get $25 in free credits!) by clicking here. Also, explore our documentation and join the Slack community to learn more about what the Grid platform can do for you.

Grid Platform