Home Networking company Strong Compute wants to speed up your ML model training – TechCrunch

Strong Compute wants to speed up your ML model training – TechCrunch


Training neural networks takes a long time, even with the fastest and most expensive accelerators on the market. It’s perhaps unsurprising, then, that a number of startups are looking to speed up the process at the software level and remove some of the current bottlenecks in the training process. For Strong Compute, a Sydney, Australia-based startup that was recently accepted into Y Combinator’s Winter ’22 class, it’s about removing those inefficiencies from the training process. By doing so, the team claims they can speed up the training process by 100 times or more.

“PyTorch is beautiful, and so is TensorFlow. These toolkits are amazing, but their simplicity — and ease of implementation — comes at the expense of inefficient things under the hood,” said CEO and Founder of Strong Compute, Ben Sand, who previously co-founded the company AR Meta (before Facebook used that name).

Although some companies focus on optimizing the models themselves, and Strong Compute also does so if requested by its customers, Sand noted that this could compromise the results. What the team is focusing on instead is all about the model. This can be a slow data pipeline or pre-computing a large number of values ​​before training begins. Sand also noted that the company has optimized some of the often-used libraries for data augmentation.

The company also recently hired Richard Pruss, a former Cisco principal engineer, to focus on removing network bottlenecks in the training pipeline, which can quickly lead to a lot of latency. But, of course, hardware can also make a big difference, which is why Strong Compute also works with its customers to run models on the right platform.

“”Strong Compute cut our basic algorithm training from thirty hours to five minutes, training hundreds of terabytes of data,” said Miles Penn, CEO of MTailor, which specializes in creating custom clothing for its customers. online.” Deep learning engineers are probably the most valuable resource on this planet, and Strong Compute has enabled ours to be 10 times more productive. Iteration and experimentation time is the most important lever for ML productivity, and we were lost without Strong Compute.

Sand says the big cloud providers don’t really have an incentive to do what his company does, given that their business model relies on people using their machines for as long as possible, with which Michael Seibel, CEO of Y Combinator , agrees. “Strong Compute is targeting a serious misalignment of incentives in cloud computing, where faster results that are valued by customers are less profitable for providers,” Seibel said.

Picture credits: Strong Compute’s Ben Sand (left) and Richard Pruss (right).

Currently, the team is still providing top-notch service to its customers, although developers shouldn’t notice much of a difference as integrating its optimizations shouldn’t really change their workflow. The promise Strong Compute makes here is that it can “10x your development cycles”. For the future, the idea is to automate the process as much as possible.

“AI companies can focus on their core customer, data and algorithm where their core IP and value is, leaving all configuration and operations to Strong Compute,” Sand said. This not only gives them the rapid iteration they need to succeed, but also ensures that their developers only focus on work that adds value to the business. Today, they spend up to two-thirds of their time on complex “ML Ops” systems administration tasks, which are largely generic in AI companies and often outside their area of ​​expertise; it doesn’t make sense for it to be in-house. »

Premium: Here’s a video of our own Lucas Matney trying out the Meta 2 AR headset from Sand’s latest company in 2016.