🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Our understanding of MLOps is limited to a fragmented landscape of thought pieces, startup landing pages, & press releases. So we did interview study of ML engineers to understand common practices & challenges across organizations & applications: arxiv.org/abs/2209.09125

A great presentation yesterday by my colleague Matthias Langer from NVIDIA AI yesterday at ACM RecSys 2022 on the embedding parameter server that he and the team developed as a part of HugeCTR. Already available online here: vimeo.com/752339625/6ece…

Our NVIDIA AI blog on the ACM RecSys #recsys2022 challenge solution from this year's competition.

Our ML talk series next week features Building End-to-End Recommender Systems with Nvidia Merlin, by Ronay Ak & Benedikt from NVIDIA AI Even Oldridge 📚 Join to win 5 recsys books by Packt Publishing ⏰ Sep 27, 10AM PT | 5PM GMT 📌 aicamp.ai/event/eventdet…

I am launching a new blog -- TabularMusings 🥳 Here is the first blog post: tabularmusings.com/posts/feature-… And here is the technology I am using and the reasons for starting the blog:

3 new example notebooks just got merged! 🥳 In them I walk you through: • how to gain access to cutting-edge tabular data preprocessing to improve results and cut down development time • scale to arbitrarily large data thanks to CUDA clusters github.com/NVIDIA-Merlin/…

This talk looks fantastic. The paper by Shreya Shankar et. al. is an incredible treatise on MLOps challenges and solutions and the initial video from Hugo summarizing it leaves me very excited for tomorrow's fireside chat.

Excited to see the latest MLPerf results with Hopper H100s improving upon A100s for DLRM by over 50%! Looking forward to the next iteration when Grace Hopper systems come online! blogs.nvidia.com/blog/2022/11/0…

Merlin Dataloader is 119x faster than my own PyTorch Dataset + Dataloader combo! This is revolutionary for tabular data 🥳 Let's take a closer look at what is going on.

If you'd like to take the new Merlin Dataloader library for a spin 👉 I created a Kaggle kernel that trains a matrix factorization model to generate candidates for a live RecSys competition 😊 Please find it here: kaggle.com/code/radek1/ma…

For all my 🇺🇸 friends getting over their 🦃hangovers here's a black friday blog for you. Ever wanted to scale your recommender system to the Terabyte level. Check out this amazing work by Hao Wu and Deya Fu on the Merlin Distributed Embeddings library: developer.nvidia.com/blog/fast-tera…

An in-depth comparison of using • Petastorm • TensorFlow • Merlin Dataloader for loading data on DataBricks! Merlin is 10x - 1000x faster 🙂 Thank you so much Sandro Cavallari for this fascinating study! Read more here: andompesta.github.io/2022/11/13/eff…

What is a sure way to improve the performance of your ML solution? 👉 Ensembling And what is the easiest way to introduce diversity into your ensemble? 👉 Training with different losses You can do both at GPU speeds with Merlin 🧙‍♂️ Learn more here 👉 kaggle.com/competitions/o…

We just released a comprehensive survey paper on causal inference in recommender systems arxiv.org/pdf/2301.00910…. It covers widely-used strategies for bias mitigation, explanation, and generalization. Please check it out!

Interested in what the NVIDIA AI Merlin team has been up to? Join us tomorrow for a demo of how to build session-based recommender systems. info.nvidia.com/Recommender-Sy…

GPUs are of course amazing, but it's helpful to be able to run your code on CPU when you're iterating quickly on a small dataset. Rick Zamora (who works on NVTabular with the Merlin team) & @quasiben have worked hard to make this easy for dataframes in Dask!

Well I've managed to ruin my Spotify recommendations by using it as a sound machine while sleeping. My formerly great recs are now all mellow soundscapes and white noise. Spotify Research can you please exclude that kind of content?

Want to see session based recommendations with NVIDIA AI Merlin (github.com/NVIDIA-Merlin/…) live in action? For this year's GTC Conference we served up recs powered by Merlin. Check out the details here: blogs.nvidia.com/blog/2023/04/0…

If you watched my "Getting Started with CUDA for Python Programmers", and are ready to go even faster, then this new video is for you! Fast CUDA isn't just about having all the threads go brrrr in parallel, but also give them fast memory to play with. youtu.be/eUuGdh3nBGo

Looking to evaluate your image-based retrieval on multilingual data? Allow me to please share a new benchmark, MIRACL-VISION: • 7898 queries • 338734 images in corpus • spanning 18 languages, including low-resource and non-Latin alphabet languages (more info below)