Learn how Spotify, Walmart, Tencent and Postmates are scaling and accelerating their recommender systems on GPU during GTC 2021
By Benedikt Schifferer, Ronay Ak and Even Oldridge
A year ago at GTC2020 Jensen announced NVIDIA’s new open source framework Merlin, to accelerate deep learning recommendation systems with GPUs. Over the past year our team has worked hard to add so many cool new features including multi-GPU and multi-node ETL, an easier to use high level API, inference support for Triton Inference Server, support for session based recommendation, and our soon to be released multi-GPU dataloader. Even more exciting are the companies that have begun using Merlin in their recommender pipelines. In this post we share some of these success stories. If you’d like to learn more there are GTC 2021 talks from each of these teams showcasing the amazing work they’ve done.
How are companies using Merlin in their recommender system pipelines?
Tencent accelerated their advertisement recommendations by 7x with NVIDIA Merlin. Xiangting Kong, who is an expert engineer at Tencent, shares how they applied NVIDIA Merlin to their advertisement system and achieved 7x speed-up in comparison to the previous TensorFlow solution using the same GPU infrastructure.
Walmart scaled their dynamic pricing system to multi-GPUs. Grant Gelven, Staff Machine Learning Engineer, presents Walmart’s solution to use Merlin NVTabular, Dask and TensorFlow to scale data preprocessing and training a wide and deep model with multi-GPU support.
Spotify reduced offline model evaluation from multiple hours to minutes. Marc and Joseph, Machine Learning Engineers at Spotify, explain how they leverage RAPIDS.AI cuDF / dask_cudf to accelerate their model evaluation pipelines with GPUs.
Postmates accelerated preprocessing and training pipelines. Lance and Frank, Machine Learning Engineer at Uber AI and VP of ML at Postmates respectively, explain their approach of using RAPIDS.AI and Merlin NVTabular in their personalization and discovery features.
Interested to learn more about NVIDIA Merlin?
Early adopters have had tremendous success in applying NVIDIA Merlin to their recommendation system pipelines and gained large speed-ups. If you are interested to learn more about NVIDIA Merlin and how you can apply it to your workflows, you should not miss the following talks:
Accelerated Recommender Systems on the GPU with Merlin. Even, Sr. Manager of NVIDIA’s Recommendation System Framework Team, holds a 40 min presentation about the challenges of recommendation systems and how NVIDIA Merlin addresses them. He gives more insights about the end-to-end vision of NVIDIA Merlin and its latest features.
We will offer a 4 hours, hands-on tutorial during GTC2021. Join our NVIDIA Merlin Tutorial for free. Benedikt and Ronay, both deep learning engineers in the Merlin team, instruct an hands-on tutorial for accelerated recommender systems on GPU(s). The tutorial will teach the audience how to create end-to-end recommender systems pipelines with NVIDIA Merlin framework.
Alternatively, you can join the discussion during our Q&A with the NVIDIA Merlin team. An expert group of senior engineers and managers at NVIDIA will meet you in either a group or 1:1 setting, to get your questions answered. The expert session will deep dive into how to optimally prepare, train, and deploy recommender systems on the GPU.
Don’t miss it! Sign up NOW for GTC2021! It’s free!
If you are interested in scaling or accelerating deep learning recommendation systems, GTC2021 is the conference for you. A selection of 19 talks from NVIDIA and a variety of guest speakers talk about their experience of GPU-accelerated recommender systems. In particular, our guests sharing their real-world applications are a highlight and you should not miss them. Sign up for free for GTC2021 and explore the recommendation system track.
