Madriss Seksaoui’s Post

Check out this quantized version of Phi-3 Vision that Jambo Chen and Sisiy Gu have working through ONNX on Jetson...would love it if these all magically exported to ONNX! Let me know if anyone tries this with the TensorRT backend in ONNX Runtime. I have been using PyTorch->ONNX->TRT for deploying the VITs.

The goal of this article is to run the ONNX format Phi-3-vision quantization model on the Jetson platform and successfully infer image + text dialogue tasks. Please kindly note:ONNX Runtime does not provide a precompiled version for aarch64 + GPU, but we can get the required library files from the dusty-nv image. Thanks for Dustin Franklin 's effort

If you’re not taking advantage of quantization, you’re losing money. Quantized models have reduced memory footprint, can be ran with less resources and faster! And you can obtain all that using Hugging Face optimum. Using the optimum package you can quantize your transformer models with just a CLI command, and there are also different options designed for different device types (CPU or GPU). You can also use the same package to export your models to the classic ONNX format. And the best part: you can use pipelines to perform your inferences, getting those beautiful results that HuggingFace pipelines usually provide! If you enjoyed this, join me in exploring the world of programming, machine learning, and MLOps by hitting the following button. Let's level up together! 🔥🤖 #mlops #machinelearning

🚀 Supercharging Data Analytics with cuDF - The GPU-Accelerated Pandas As we navigate the vast ocean of data analytics, speed and efficiency are paramount. Enter cuDF, NVIDIA's brilliant contribution to the RAPIDS AI suite, often dubbed as "CUDA Pandas." It's not just a tool; it's a game-changer for data professionals. cuDF stands at the crossroads of innovation and performance, providing a Pandas-like API but with a twist - it runs on NVIDIA GPUs. This means data preparation and processing that once took minutes or hours can now be completed in seconds or milliseconds. 🕒💡 🔍 Why cuDF? ✅ Speed: Leveraging GPU acceleration, cuDF transforms data analysis tasks, offering unprecedented processing times. ✅Efficiency: Optimize hardware resources, handling large datasets with ease and agility. ✅Compatibility: Offers a familiar Pandas-like interface, making the transition smooth and inviting for Python data professionals. Whether it's crunching large datasets or performing complex computations, cuDF opens new horizons in data processing speed and efficiency. It's not just about doing things faster; it's about unlocking new possibilities in data exploration and insights generation. 🌟📊 Discover more about how cuDF is revolutionizing data analytics in this insightful video: https://lnkd.in/esrRhUj5 Have you ever used cuDF on large datasets Venkata Naga Sai Kumar Bysani, Ayush Shinde, Geetha Sagar Bonthu? #DataAnalytics #cuDF #GPUProcessing #DataScience #RAPIDSAI #Innovation

#grok-1 model weights by #elonmusk #xAI is released: ➡Started with grok0 with 33 billion parameters ➡Performance similar to llama2 70 billion model. ➡63.2% on the HumanEval coding task and 73% on MMLU which is better than GPT3 but less than GPT4 and claude2 ➡Its good in math and coding ➡Main challenges was handling distributed model training using GPUs because GPUs might randomly fail during training. To handle the infrastructure effectively they used Rust. On this note please check out my book. https://lnkd.in/gFNx6Z-V ➡Multi GPU machine is required to test the model. Model card: https://lnkd.in/gpG5uh6S Github: https://lnkd.in/gj_qtkC2 Blog post: https://x.ai/blog/grok

Excited to share that I've earned the "Fundamentals of Accelerated Computing with CUDA C/C++" certification from NVIDIA! - Deep Learning Institute. With this certification, I turned slow CPU-based programs into a super-fast GPU-powered applications. By using CUDA GPU programming, I boosted my code's speed by a staggering 14 times! Throughout the workshop, I sharpened my skills in parallel programming, learned how to optimize threads and blocks for maximum performance, and leveraged techniques like unified memory sharing and prefetching data to further enhance speed and efficiency. Can't wait to apply these advanced CUDA strategies to solve even more complex problems! #nvidia #cuda #gpucomputing

**The Ultimate Pandas Power-Up!** Exciting Performance Boost with RAPIDS cuDF! I recently explored RAPIDS cuDF's pandas accelerator mode and was blown away by the speed gains. You can now supercharge your pandas workflows with GPU acceleration and zero code changes! Key Takeaways: Setup: Verified NVIDIA GPU availability and imported necessary libraries. Data Loading: Downloaded a substantial NYC parking violations dataset. CPU vs. GPU: Pandas on CPU: Traditional analysis took several seconds. cuDF on GPU: The same operations finished in milliseconds! Seamless Integration: Created visualizations using Plotly and enjoyed smooth third-party library integrations with cuDF. Why This Matters: With cudf.pandas, you can drastically reduce computation time for your data analysis tasks, leveraging the power of GPUs without altering your existing code. It’s a game-changer for data scientists and analysts! Curious to dive deeper? Head over to RAPIDS cuDF and get ready to supercharge your data processing! Check out my repository for a detailed performance comparison: Pandas to cuDF Performance Comparison https://lnkd.in/gXWxXBmG #DataScience #MachineLearning #AI #GPU #RAPIDS #cuDF #Pandas #BigData #Python#CUDA

😎 Help the NVIDIA stock rising to the moon: run your typical Data Science on GPUs OK ... NVIDIA might to need us, humble data scientists using Pandas, to sell their golden tickets. Yet ... If you get access to GPUs, you could speed up your Pandas analysis by up to 100x (depending on the operations and GPU, of course). RapidsAI is building cuDF (cuda DataFrame 😉) to help you run pandas on GPU steroids. Bonus point, you can load it on Colab with a simple magic command. Learn how in my new YouTube video: https://lnkd.in/ejcfSpxA We'll dive together into cuDF and benchmark a few data wrangling workflows in a Colab Notebook. See you there 👋🏽. #machinelearning #ai #cuda #data #python

At Kexxu we have a competition who can make the fastest hand-written neural network. Some of the libraries we're trying out: - BLIS (faster BLAS for C++) - MatX (by Nvidia, numpy on thr GPU in C++) - WebGPU (C++ and Rust, on GPU and huge compatibility, best chance to get GPU acceleration on raspbarey pi, phones, etc) - Hand written SIMD in C++ (my current version) Anybody know some other things to try?

MLS_09 Automatic Differentiation and Calculus for Machine Learning using JAX => JAX is NumPy on the CPU, GPU, and TPU, with great automatic differentiation framework developed by Google for ML. https://lnkd.in/gxDUQx4T