Everyone knows that great visuals are worth a thousand words. The team at Prezi, a Visual Communications software company, practices this insight with Prezi presentations that combine images and text in highly dynamic presentations.
Prezi took part in the embracing Face Expert Support program to fully utilize the possibilities of modern machine learning. Over the past few months, the hugging face has been supporting Prezi in integrating smaller, more efficient open source models into ML workflows. This collaboration began at a perfect time as multimodal models become increasingly competent.
We recently sat down with Máté Börcsök, a backend engineer at Prezi, to talk about our experiences in the professional support program. In this short video, Máté introduces us to some of our machine learning tasks and works with our team to share our experiences through our expert support program.
https://www.youtube.com/watch?v=pm6d0troibi
If you want to accelerate your machine learning roadmap with the help of experts, as Máté and his team did, visit hf.co/support to request a detailed and quote for our expert support program.
Transcript with additional details:
introduction
My name is Máté and I am a backend engineer at Prezi, an online presentation tool that will bring your ideas to life.
How can the HF Expert Support Program help you build AI?
Prezi’s flagship AI product is Prezi AI. This helps users create presentations faster and faster. Users start by providing prompts and explanations for the presentation they want to create. The system will then automatically create a draft presentation and create a draft presentation to start. This is a complex system that invokes different services and builds the structure of the presentation using a closed model and various asset provider services.
When we joined the program we already had a version of this system and our experts reviewed the flow and suggested improvements. The pipeline includes a search system to find the right assets (images and text) for each unique presentation. In this context, the important advice was to add, for example, an open source re-ranker model to the system. This makes the best images and text for your presentation cheaper, faster and better than LLM.
The use cases are inherently multimodal, as the presentation combines images and text. Many models are released each week, but our experts can help us get through the hype and understand which models will help us and which ones will not. This saves a lot of time as you solve your own challenges using a combination of vision models, text models and vision language models (VLM). Multimodal machine learning is challenging and the guidance is truly appreciated. We are not machine learning engineers, we learn this together along the way.
What is your favorite feature of inference endpoints?
We highly recommend checking the endpoint model catalog. This is a curated list of models that work well on inference endpoints and require zero configuration. I love that the endpoints can be set up so that they fall asleep in a few minutes. It also supports single and quad A100 instances required for some models. It’s also easy to update the model. Inference endpoints can be used to deploy the latest version or roll back to an older version with just one click using Git Hash. Using them was extremely convenient as none of these features are easily available on AWS. Even if the models are not in the catalog yet, it’s relatively easy to get them to work. It was easy for me at least. Our experts were supporting us.
Which teams will benefit most from professional support?
The embracing face partnership has opened the door to machine learning for us. Our dedicated experts provide access to a community of machine learning experts who can give feedback on our wildest questions. As I said, we are not machine learning engineers. Our experts share best practices and cutting-edge models for embedding, reranking, and object detection, and guide us to work on the right things to show how to fine-tune new vision language models to collect and curate data. These are mainly things we can do, but his guidance gives a huge speedup and keeps us focused on tasks that make sense to our users.
With the HE Expert Support program, we have built a world-class team to help our customers make ML solutions better faster. Our experts answer questions and find solutions as needed on our machine learning journey from research to production. Visit hf.co/support to find out more and request a quote.
