Corey J Nolet retweeted
We have truly built something great here. From composing machine learning algorithms with a ready-to-go toolbox of building blocks to drop-accelerators for existing libraries. We are just getting started!
The idea of RAFT - a box of mathematical building blocks, fully optimized for GPU, has been core to the RAPIDS vision from the start. RAFT is much, much more than an today idea. @cjnolet and team have provided the world with something amazing. developer.nvidia.com/blog/re…
Corey J Nolet retweeted
The idea of RAFT - a box of mathematical building blocks, fully optimized for GPU, has been core to the RAPIDS vision from the start. RAFT is much, much more than an today idea. @cjnolet and team have provided the world with something amazing. developer.nvidia.com/blog/re…
We have truly built something great here. From composing machine learning algorithms with a ready-to-go toolbox of building blocks to drop-accelerators for existing libraries. We are just getting started!
The idea of RAFT - a box of mathematical building blocks, fully optimized for GPU, has been core to the RAPIDS vision from the start. RAFT is much, much more than an today idea. @cjnolet and team have provided the world with something amazing. developer.nvidia.com/blog/re…
Corey J Nolet retweeted
#Milvus 2.3 Beta delivers
⏩ GPU acceleration + RAFT-based integration
🔭 Range search for precision in complex searches
🦋 CDC to deliver changes to downstream systems
🗺 mmap for performance improvement
🔺Upserts (update or insert)
Details bit.ly/3JWdKZK 🧵
Corey J Nolet retweeted
Nvidia #GPU support brings the ability to support heterogeneous computing, which can significantly accelerate specialized workloads.
This new addition allows users to expect faster and more efficient #VectorSearch, ultimately improving productivity and performance.
The #raft library by @rapidsai contains a rich catalog of highly reusable computational patterns for composing #MachineLearning and #DataAnalytics applications.
RAFT also contains state-of-the-art implementations of new approximate nearest neighbors algorithms on the GPU!
Spend more time designing your app & less time worrying about bottlenecks.
Read a quick primer on RAPIDS RAFT's library of building blocks for #ML & #DataAnalysis: nvda.ws/3FKxqNU
Corey J Nolet retweeted
The Merlin team is always building something exciting! medium.com/nvidia-merlin/bui…
Corey J Nolet retweeted
github.com/rapidsai/raft is probably the best place to look 👀
Corey J Nolet retweeted
The integration was done into Knowhere - Milvus' core vector search engine - as a new index type. You can check it out under src/index/ivf_raft.
github.com/milvus-io/knowher…
Corey J Nolet retweeted
We've got so much stuff in our latest release, including cuML that can run on CPU or GPU. Check it out! medium.com/rapids-ai/rapids-…
Another great example of how RAPIDS is making data science more productive and enabling methods that were once intractable. #neighborhoodmethods #weallneedbalance
Tip the scales with more speed and balance your datasets with cuML and imbalanced-learn resampling @becker_data @dante_dgd @cjnolet medium.com/rapids-ai/faster-…
Corey J Nolet retweeted
Intern position for computer architect role at NVIDIA UK Reading:
nvidia.wd5.myworkdayjobs.com…
Corey J Nolet retweeted
Learn about the newest release of CUDA and its exciting features and capabilities in this webinar and live Q&A! bit.ly/3Po0Fd5
Corey J Nolet retweeted
CUDA12 is available! It supports DPX for Hopper and older GPUs by emulation.
Old-school CUDA devs will find DPX a natural evolution of the video SIMD instructions included >12 years ago in Fermi GPUs.
DPX uses same Math API, legacy codes will be hw acc:
docs.nvidia.com/cuda/cuda-c-…
Corey J Nolet retweeted
Wondering what I've been up to lately?
I'm excited to share my new free course:
Deep Learning Fundamentals
-- Learning Deep Learning Using a Modern Open Source Stack!
I hope you’ll find it useful! Check it out here:
👉 lightning.ai/pages/courses/d…
Let me know what you think!
Corey J Nolet retweeted
Amazing work by @cjnolet and @becker_data! Hours to seconds! You owe it to yourself to check out this blog. teapaw.org/NVIDIAAIDev/status/160…
Waiting hours for HDBSCAN soft clustering to finish processing a few hundred thousand records? The RAPIDS cuML library can do HDBSCAN soft clustering in seconds rather than hours. Get the code and latest benchmarks: nvda.ws/3BfnMQV
HDBSCAN is one of the few density based clustering algorithms that enable out of sample inference in addition to fuzzy clustering. Doing this at the speed of light is absolutely game changing.
Amazing work by @cjnolet and @becker_data! Seconds to hours! You owe it to yourself to check out this blog.
