Free-form floor plan design using differentiable Voronoi diagram (Score: 151+ in 10 hours)
Link: https://readhacker.news/s/6f477
Comments: https://readhacker.news/c/6f477
We fine-tuned Llama 405B on AMD GPUs (🔥 Score: 152+ in 3 hours)
Link: https://readhacker.news/s/6f5bb
Comments: https://readhacker.news/c/6f5bb
Hey HN, we recently fine-tuned the llama3.1 405B model on 8xAMD MI300x GPUs using JAX instead of PyTorch. JAX's advanced sharding APIs allowed us to achieve great performance. Check out our blog post to learn about the cool sharding tricks we used. We've also open-sourced the code: https://github.com/felafax/felafax
We're a small startup building AI infra for fine-tuning and serving LLMs on non-NVIDIA hardware (TPUs, AMD, Trainium).
Problem: Many companies are trying to get PyTorch working on AMD GPUs, but we believe this is a treacherous path. PyTorch is deeply intertwined with the NVIDIA ecosystem in a lot of ways (e.g., `torch.cuda` or scaled_dot_product_attention is an NVIDIA CUDA kernel exposed as a PyTorch function). So, to get PyTorch code running on non-NVIDIA hardware, there's a lot of "de-NVIDIAfying" that needs to be done.
Solution: We believe JAX is a better fit for non-NVIDIA hardware. In JAX, ML model code compiles to hardware-independent HLO graphs, which are then optimized by the XLA compiler before hardware-specific optimization. This clean separation allowed us to run the same LLaMA3 JAX code both on Google TPUs and AMD GPUs with no changes.
Our strategy as a company is to invest upfront in porting models to JAX, then leverage its framework and XLA kernels to extract maximum performance from non-NVIDIA backends. This is why we first ported Llama 3.1 from PyTorch to JAX, and now the same JAX model works great on TPUs and runs perfectly on AMD GPUs.
We'd love to hear your thoughts on our vision and repo!
A terrible way to jump into colocating your own stuff (Score: 150+ in 17 hours)
Link: https://readhacker.news/s/6f2xF
Comments: https://readhacker.news/c/6f2xF
Foundations: Why Britain Has Stagnated (❄️ Score: 168+ in 3 days)
Link: https://readhacker.news/s/6eTs6
Comments: https://readhacker.news/c/6eTs6
Alan Turing's 1950 manual for the Mark I electronic computer [pdf] (Score: 150+ in 16 hours)
Link: https://readhacker.news/s/6f2tv
Comments: https://readhacker.news/c/6f2tv
Tesla Transport Protocol over Ethernet (TTPoE) (Score: 151+ in 16 hours)
Link: https://readhacker.news/s/6f2ei
Comments: https://readhacker.news/c/6f2ei
Coffee Stats – Maximize Caffeine Intake and Get to Bed at Night (Score: 150+ in 19 hours)
Link: https://readhacker.news/s/6eZGk
Comments: https://readhacker.news/c/6eZGk
I Designed a Dieter Rams Inspired iPhone Dock (Score: 167+ in 7 hours)
Link: https://readhacker.news/s/6f2Nd
Comments: https://readhacker.news/c/6f2Nd
Things I've learned serving on the board of the Python Software Foundation (❄️ Score: 151+ in 4 days)
Link: https://readhacker.news/s/6eM95
Comments: https://readhacker.news/c/6eM95
How the iPhone 16's electrically-released adhesive works (🔥 Score: 152+ in 3 hours)
Link: https://readhacker.news/s/6f2Jm
Comments: https://readhacker.news/c/6f2Jm
Show HN: Time Flies (Score: 152+ in 1 day)
Link: https://readhacker.news/s/6eWuM
Comments: https://readhacker.news/c/6eWuM
A visualization of the passage of time using flies. Written in JavaScript with some HTML & CSS in one index.html.
What's inside the QR code menu at this cafe? (🔥 Score: 150+ in 1 hour)
Link: https://readhacker.news/s/6f2xi
Comments: https://readhacker.news/c/6f2xi
London saw a surprising benefit to ultra-low emissions zone: More active kids (Score: 150+ in 6 hours)
Link: https://readhacker.news/s/6f22w
Comments: https://readhacker.news/c/6f22w
Apple adds feature that tries to "autocorrect" file extensions, and fails (Score: 153+ in 9 hours)
Link: https://readhacker.news/s/6eZm8
Comments: https://readhacker.news/c/6eZm8
Mozilla exits the Fediverse and will shutter its Mastodon server in December (❄️ Score: 150+ in 5 days)
Link: https://readhacker.news/s/6eJrB
Comments: https://readhacker.news/c/6eJrB
What, Me Worry? The Art and Humor of Mad Magazine (Score: 151+ in 9 hours)
Link: https://readhacker.news/s/6f4bm
Comments: https://readhacker.news/c/6f4bm
In 1870, Lord Rayleigh used oil and water to calculate the size of molecules (Score: 154+ in 4 hours)
Link: https://readhacker.news/s/6f4Hv
Comments: https://readhacker.news/c/6f4Hv
Intel's Redwood Cove: Baby Steps Are Still Steps (Score: 151+ in 1 day)
Link: https://readhacker.news/s/6eZbx
Comments: https://readhacker.news/c/6eZbx
Show HN: I Wrote a Book on Java (🔥 Score: 151+ in 2 hours)
Link: https://readhacker.news/c/6f4FK
https://www.manning.com/books/data-oriented-programming-in-j...
This book is a distillation of everything I’ve learned about what effective development looks like in Java (so far!). It's about how to organize programs around data "as plain data" and the surprisingly benefits that emerge when we do. Programs that are built around the data they manage tend to be simpler, smaller, and significantly easier understand.
Java has changed radically over the last several years. It has picked up all kinds of new language features which support data oriented programming (records, pattern matching, `with` expressions, sum and product types). However, this is not a book about tools. No amount of studying a screw-driver will teach you how to build a house. This book focuses on house building. We'll pick out a plot of land, lay a foundation, and build upon it house that can weather any storm.
DoP is based around a very simple idea, and one people have been rediscovering since the dawn of computing, "representation is the essence of programming." When we do a really good job of capturing the data in our domain, the rest of the system tends to fall into place in a way which can feel like it’s writing itself.
That's my elevator pitch! The book is currently in early access. I hope you check it out. I'd love to hear your feedback.
You can get 50% off (thru October 9th) with code `mlkiehl`
https://www.manning.com/books/data-oriented-programming-in-j...
The Intelligence Age (🔥 Score: 156+ in 2 hours)
Link: https://readhacker.news/s/6f4i9
Comments: https://readhacker.news/c/6f4i9
Cloudflare's new marketplace lets websites charge AI bots for scraping (🔥 Score: 154+ in 3 hours)
Link: https://readhacker.news/s/6f3zH
Comments: https://readhacker.news/c/6f3zH
GPU Puzzles (❄️ Score: 154+ in 5 days)
Link: https://readhacker.news/s/6eLGW
Comments: https://readhacker.news/c/6eLGW
Mathematicians discover new class of shape seen throughout nature (Score: 150+ in 19 hours)
Link: https://readhacker.news/s/6eZgP
Comments: https://readhacker.news/c/6eZgP
Cloudflare Is Breaking My SVGs? (Score: 150+ in 1 day)
Link: https://readhacker.news/s/6eXXh
Comments: https://readhacker.news/c/6eXXh
Valve is testing ARM64 support for popular games (Score: 150+ in 13 hours)
Link: https://readhacker.news/s/6eZCS
Comments: https://readhacker.news/c/6eZCS
Low Cost CO2 Sensors Comparison: Photo-Acoustic vs. NDIR (Score: 150+ in 8 hours)
Link: https://readhacker.news/s/6eZZm
Comments: https://readhacker.news/c/6eZZm
The elite's war on remote work has nothing to do with productivity (🔥 Score: 158+ in 1 hour)
Link: https://readhacker.news/s/6f2xs
Comments: https://readhacker.news/c/6f2xs
Desktop Windowing on Android Tablets (Score: 151+ in 9 hours)
Link: https://readhacker.news/s/6eZqM
Comments: https://readhacker.news/c/6eZqM
It's probably time to rethink “Building in public” (Score: 150+ in 9 hours)
Link: https://readhacker.news/s/6eZdB
Comments: https://readhacker.news/c/6eZdB
Rawdrawandroid – Build Android apps without any Java, in C and Make (Score: 151+ in 12 hours)
Link: https://readhacker.news/s/6eYK7
Comments: https://readhacker.news/c/6eYK7