Grid.ai today announced the general availability of Grid, a new platform that enables researchers and data scientists to train AI models in the cloud. The company says that Grid enables development and training “at scale” without requiring advanced skills in machine learning engineering.
Machine learning practitioners can run into challenges when scaling AI workloads because of the infrastructure needed to train and deploy into production. Moreover, this infrastructure can be expensive to maintain. A Synced study estimated that the University of Washington’s Grover fake news detection model cost $25,000 to train in about two weeks and OpenAI reportedly racked up a whopping $12 million to train its GPT-3 language model.
At a high level, Grid offers an interface for training models on GPUs, processors, and more. It’s a web app with a command line interface that optimizes datasets to work at the level needed for production and “cutting-edge” research. Grid computes in real time to make it easier to quantify the R&D efforts of AI projects. Additionally, the platform provides access to Jupyter notebooks, a way for data scientists to bundle their answers with the Python code that produced it.
To use Grid, users simply clone a project from their GitHub repositories and make minor code changes. Data and artifacts stay on their infrastructure — Grid only orchestrates. And experiment artifacts, like weights and logs, are automatically saved to the cloud so that they can be accessed during or after a training run.
Grid will be available starting April 13 in three plans: individual, team, and enterprise. The individual plan is the cost of the cloud processing plus 20%, while the team plan is $1,000 a month in addition to the cloud costs and 15%.
The launch of Grid comes after its namesake company, Grid.ai, emerged from stealth in October 2018 with $18.6 million in venture capital backing. Cofounded by William Falcon, the creator of PyTorch Lightning, one of the fastest-growing machine learning frameworks in the world, Grid.ai’s mission is to reduce the distance between deep learning research and its practice in real-life businesses.
Grid leverages PyTorch Lightning to decouple the code required to define a full deep learning model from the code required to run it on hardware. Falcon says the main goal is to help companies focus on delivering value instead of worrying about the machines they’re running or clusters.
“Think back to when electricity came about. Only a few had access to it until the power grid made it available to everyone,” Falcon told VentureBeat via email. “That’s the goal of Grid.ai, to democratize access by removing the need to be an expert engineer. ML practitioners should be focused on delivering value through models and data, not on becoming expert engineers.”
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.
Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:
- up-to-date information on the subjects of interest to you
- our newsletters
- gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
- networking features, and more
Become a member