Weights & Biases, a leading platform in AI and machine learning development, has recently secured a significant investment from notable figures in the tech industry, including former GitHub CEO Nat Friedman and ex-Y Combinator partner Daniel Gross. Along with existing investors such as Coatue, Insight Partners, Felicis, Bond, BloombergBeta, and Sapphire, they have contributed $50 million in a funding round that values the company at $1.25 billion. This fresh capital raises Weights & Biases' total funding to $250 million and comes just as the startup prepares to launch "Prompts," a new product aimed at helping users monitor and assess the performance of large language models (LLMs), akin to OpenAI's GPT-4.
While the latest $50 million funding round is smaller compared to the company’s prior Series C round, which raised around $135 million, Lavanya Shukla, VP of Growth at Weights & Biases, views this investment as strategic. In a recent email interview, she emphasized, “Empowering employees with machine learning tools is essential for CTOs and their teams. By addressing critical aspects like testing, security, and reliability, Weights & Biases plays a vital role in developing successful machine learning models.”
Co-founders Lukas Biewald and Chris Van Pelt launched Weights & Biases in 2017 after years of creating tools for machine learning engineers and data scientists. They previously established Figure Eight, formerly known as CrowdFlower, to facilitate crowdworkers in labeling training data for machine learning algorithms (Figure Eight was acquired by Appen in 2019 for $175 million). “They recognized a significant gap: machine learning practitioners lacked a reliable system to document their experiments,” Shukla explained. “Instead of using effective logging methods, they relied on poorly organized spreadsheets and deteriorating screenshots.”
To tackle this issue, Biewald and Van Pelt partnered with Shawn Lewis, a former Google developer, to create the minimum viable product (MVP) for Weights & Biases, which includes workflows that streamline the machine learning development lifecycle. As a prominent contender in the MLOps (machine learning operations) landscape, Weights & Biases enables data scientists to build machine learning models and execute repeatable, automated workflows for deployment. With the burgeoning demand for AI technologies, the MLOps sector is projected to reach a valuation of $23.1 billion by 2023, according to Allied Market Research.
Numerous MLOps platforms are emerging, including names like Seldon, FedML, Qwak, Galileo, Striveworks, Arize, Comet, and Tecton, alongside established services from providers such as Azure, AWS, and Google Cloud. However, Shukla asserts that Weights & Biases differentiates itself through its user-centric approach. “All our products are co-designed with partners and customers to align with their needs,” she said. The platform also emphasizes tools that analyze datasets for model training, enabling users to identify potential issues—such as biases and the inclusion of personally identifiable information—before production.
Shukla stated, “Weights & Biases stands out as the premier machine learning platform that empowers developers to build superior models more efficiently. Our lightweight, interoperable tools streamline experiment tracking, dataset versioning, model evaluation, regression detection, and collaborative sharing. This approach allows machine learning engineers to iterate on their pipelines confidently, knowing their datasets and models are meticulously tracked in a reliable system.”
The first-mover advantage cannot be overlooked for Weights & Biases. The platform is integrated across more than 20,000 open-source repositories and has been referenced in numerous machine learning academic research papers. It is also the go-to tool for top-tier generative AI model developers like OpenAI, Aleph Alpha, Cohere, Anthropic, and Hugging Face. Shukla noted, “OpenAI relies on Weights & Biases for training all its models. With hundreds of employees managing thousands of experiments, the platform is essential for testing, identifying issues, and quickly debugging their models. It has also accelerated training runs for GPT-4.”
Beyond this influential cohort, Weights & Biases has amassed an impressive user base of 700,000—up from just 100,000 in 2021—with over 1,000 paying customers. The team has expanded to over 200 members, primarily located in San Francisco.
With the launch of Prompts, Weights & Biases aims to broaden its customer base. This upcoming product will enable users to analyze LLM outputs and fine-tune the models themselves. “While LLMs may reduce the number of people required to train models, they will increase the need for professionals who can fine-tune, interact with, and create applications based on these models,” Shukla explained. “Prompts will cater to a new class of users and revolutionize how large labs approach machine learning development. Beyond prompt engineers and fine-tuners, our tools will support researchers and organizations in enhancing their unique internal models.”
Moving forward, Weights & Biases is poised to continue expanding its suite of MLOps products.
Ranking
