Microsoft Azure Integrating Hugging Face for Machine Learning Models

  • Hugging Face, a popular data science tool worth $2 billion, launched an Azure machine learning tool.
  • It gives Azure another way to win over customers that might be mostly working with other providers.
  • Hugging Face has to grow into its new $2 billion valuation, and this gives it one way to do that.

As the competition for capturing the machine learning industry heats up, Microsoft is turning to a popular $2 billion startup to get an edge over rivals.

The creators of Azure are rolling out an integration with Hugging Face, a popular data science startup that hosts some of the most-used machine learning models, to gain a new route into companies and grow business. The startup recently raised $100 million at a $2 billion valuation led by Lux Capital in a highly competitive funding round, with Addition and Sequoia participating.

Microsoft is betting that Endpoints, Hugging Face’s new integration, will help drastically simplify the time required to get machine learning models into place. The majority of efforts in machine learning die before seeing the light of day due to the number of people involved — which Hugging Face is trying to drop to as small a number as possible by making it easy for a single person to share a model across the organization.

“Most machine learning projects never make it into production,” Hugging Face head of product Jeff Boudier told Insider. “I think that’s one of the big contributing pain points is there’s this disconnect between the data science team, the infrastructure team, the compliance team, and the compute resources. To give that power to a single user, they need to be using tools that are already compliant and pass security checks and comply with processes.”

Even if many of those models don’t end up fully adopted in products, getting them on Azure in even a small capacity helps Microsoft gain credibility among machine learning experts. It’s an increasingly valuable user base that companies like Snowflake, which bought a small machine learning startup for $800 millionare increasingly chasing.

Boudier said even companies that rely on other cloud providers like Amazon Web Services could quickly deploy machine learning models on Azure that other services can call upon. A search engine running on a different cloud, for example, could quickly call upon a model trained and stored in Azure to serve up personalized results.

The strategic relationship offers Hugging Face a way to formalize relationships with many companies that use Hugging Face in a more ad-hoc manner with models stored by the startup. Microsoft, meanwhile, gets a wedge within data science teams that are eager to get models out the door and don’t want to deal with the friction that usually comes with it.

While it’s natural a company like Hugging Face would want to expand its overall base with an integration like this, these moves are rarely made in a vacuum. Microsoft has an incentive to grow more popular in the machine learning community as companies like Amazon and Snowflake aggressively invest in competing products. Hugging Face offers a strategic opportunity for the cloud giant, the kind which can sometimes becomes more robust down the line.

“This partnership furthers our collective desire to make it easier for enterprise data science teams to get started with Hugging Face on Azure,” said Jamal Robinson, Senior Director of Business Development for AI Platform & Services. “Today’s announcement gives Azure customers the most comprehensive Hugging Face platform on the cloud and furthers our goal of making state of the art NLP, audio and computer vision models more readily accessible to Azure’s software developers, citizen data scientist and advanced machine learning practitioners.”

Hugging Face is used by machine learning practitioners to download and run popular machine learning models like OpenAI’s GPT-2 and Google’s BERT. Both make it easy to quickly begin analyzing large blocks of text without spending time and resources training their machine learning models. Hugging Face also has thousands of models spanning a variety of use cases like computer vision and audio.

Leave a Comment

Your email address will not be published. Required fields are marked *