This text is a cross-post from an originally published post on April 6, 2023 in Snorkel’s blog, by Friea Berg .
As OpenAI releases GPT-4 and Google debuts Bard in beta, enterprises all over the world are excited to leverage the ability of foundation models. As that excitement builds, so does the conclusion that the majority firms and organizations will not be equipped to properly make the most of foundation models.
Foundation models pose a novel set of challenges for enterprises. Their larger-than-ever size makes them difficult and expensive for firms to host themselves, and using off-the-shelf FMs for production use cases could mean poor performance or substantial governance and compliance risks.
Snorkel AI bridges the gap between foundation models and practical enterprise use cases and has yielded impressive results for AI innovators like Pixability. We’re teaming with Hugging Face, best known for its enormous repository of ready-to-use open-source models, to supply enterprises with much more flexibility and alternative as they develop AI applications.
Foundation models in Snorkel Flow
The Snorkel Flow development platform enables users to adapt foundation models for his or her specific use cases. Application development begins by inspecting the predictions of a specific foundation model “out of the box” on their data. These predictions grow to be an initial version of coaching labels for those data points. Snorkel Flow helps users to discover error modes in that model and proper them efficiently via programmatic labeling, which may include updating training labels with heuristics or prompts. The bottom foundation model can then be fine-tuned on the updated labels and evaluated once more, with this iterative “detect and proper” process continuing until the adapted foundation model is sufficiently top quality to deploy.
Hugging Face helps enable this powerful development process by making greater than 150,000 open-source models immediately available from a single source. Lots of those models are specialized on domain-specific data, just like the BioBERT and SciBERT models used to display how ML could be used to identify opposed drug events. One – or higher yet, multiple – specialized base models may give users a jump-start on initial predictions, prompts for improving labels, or fine-tuning a final model for deployment.
How does Hugging Face help?
Snorkel AI’s partnership with Hugging Face supercharges Snorkel Flow’s foundation model capabilities. Initially we only made a small variety of foundation models available. Each required a dedicated service, making it prohibitively expensive and difficult for us to supply enterprises the pliability to capitalize on the rapidly growing number of models available. Adopting Hugging Face’s Inference Endpoint service enabled us to expand the variety of foundation models our users could tap into while keeping costs manageable.
Hugging Face’s service allows users to create a model API in a number of clicks and start using it immediately. Crucially, the brand new service has “pause and resume” capabilities that allow us to activate a model API when a client needs it, and put it to sleep once they don’t.
“We were pleasantly surprised to see how straightforward Hugging Face Inference Endpoint service was to establish.. All of the configuration options were pretty self-explanatory, but we also had access to all the choices we wanted by way of what cloud to run on, what security level we wanted, etc.”
– Snorkel CTO and Co-founder Braden Hancock
How does this help Snorkel customers?
Few enterprises have the resources to coach their very own foundation models from scratch. While many could have the in-house expertise to fine-tune their very own version of a foundation model, they might struggle to assemble the quantity of knowledge needed for that task. Snorkel’s data-centric platform for developing foundation models and alignment with leading industry innovators like Hugging Face help put the ability of foundation models at our users’ fingertips.
“With Snorkel AI and Hugging Face Inference Endpoints, firms will speed up their data-centric AI applications with open source on the core. Machine Learning is becoming the default way of constructing technology, and constructing from open source allows firms to construct the precise solution for his or her use case and take control of the experience they provide to their customers. We’re excited to see Snorkel AI enable automated data labeling for the enterprise constructing from open-source Hugging Face models and Inference Endpoints, our machine learning production service.”
Clement Delangue, co-founder and CEO, Hugging Face
Conclusion
Together, Snorkel and Hugging Face make it easier than ever for big firms, government agencies, and AI innovators to get value from foundation models. The flexibility to make use of Hugging Face’s comprehensive hub of foundation models implies that users can pick the models that best align with their business needs without having to take a position within the resources required to coach them. This integration is a big step forward in making foundation models more accessible to enterprises all over the world.
If you happen to’re fascinated with Hugging Face Inference Endpoints on your company, please contact us here – our team will contact you to debate your requirements!
