iiinova — where your digital self keeps evolving

Paulo Valente @polvalente.social · 02/06/2025

It was awesome to finally be able to showcase this! Quite a few months of work were put into getting this to work. Project and demo code can be found at: github.com/vittoriabitt...

github.com

GitHub - vittoriabitton/nx_hailo

Contribute to vittoriabitton/nx_hailo development by creating an account on GitHub.

0106

Paulo Valente @polvalente.social · 11/05/2025

There's still a bit of ground to cover before we can run Axon NNs without some very manual configuration, but this example showcases a simple Nx computation that can be distributed with Handoff: hexdocs.pm/handoff/nx_p...

hexdocs.pm

Nx Pipeline Example — Handoff v0.1.0

060

Paulo Valente @polvalente.social · 11/05/2025

It's also important to recognize that this is a problem I've started trying to solve while still with @dockyard.com, during our @mozilla.ai grant. After many months, the work is now finally paying off!

160

Paulo Valente @polvalente.social · 11/05/2025

I've just released Handoff, a new Elixir library for distributed function graph execution. hexdocs.pm/handoff Among other things, the library enables distributed Nx computations (anyone said distributed inference?) A huge part of this was sponsored by tvlabs.ai!

hexdocs.pm

Handoff v0.1.0 — Documentation

25715

Reposted by Paulo Valente

Dave Lucia @davelucia.com · 08/04/2025

And now I'm excited to announce that I was able to get the project transferred to @theerlef.bsky.social for a long-term home! TV Labs has already merged a number of issues, and even added a huge upgrade to grading thanks to @polvalente.social! #ElixirLang github.com/erlef/elixir...

github.com

feat: add self-grading Smart Cell by polvalente · Pull Request #84 · erlef/elixir-secure-coding

Also adds self-evaluation and updates for owasp module.

1183

Reposted by Paulo Valente

Code BEAM @codebeam.bsky.social · 31/03/2025

Scale Edge AI with Elixir & Nerves! Join Vittoria Bitton & @polvalente.social to explore ML on Raspberry Pi + Hailo HAT for real-time signal processing. Don't miss it! codebeamstockholm.com/talks/Scalin...

012

Reposted by Paulo Valente

Maggie Tate @maggietate.bsky.social · 05/03/2025

Me and the doggo having a grand time editing the next BeamRadio with @polvalente.social @lawik.bsky.social @ektastrophe.bsky.social Dropping later this week! #ElixirLang

Laptop with video editing of #BeamRadio and a big white doggy
on a yellow quilt to the right.

0103

Paulo Valente @polvalente.social · 04/02/2025

I'm thrilled to share that yesterday was my first day working at TV Labs! Between embedded systems, DSP, AI and regular web development, I expect a lot of cool stuff to come out of this in the near future. I'm still working on Nx as well, and that won't change anytime soon :)

1275

Paulo Valente @polvalente.social · 22/12/2024

I also prefer if, generally. But it's important to keep in mind that refactoring one into the other is not always exactly behavior-equivalent.

030

Paulo Valente @polvalente.social · 22/12/2024

That's not the same as an if because it's strictly true/false. if is more permissive, considering nil as falsy as well, and all other non booleans as true

230

Reposted by Paulo Valente

Brian Cardarella @bcardarella.bsky.social · 16/12/2024

I'm happy to announce that @dockyard.com is now offering Enterprise Support for @elixir-lang.org and Phoenix projects. We've been working on this service tier for a while. If your company is interested in this please reach out dockyard.com/services/eli... #ElixirLang

dockyard.com

Elixir Enterprise Support

Get the support you need to keep your Elixir app ahead of the competition.

0338

Paulo Valente @polvalente.social · 08/12/2024

Just came back from my last trip of 2024. Pretty intense and productive year in #ElixirLang and Nx talks: 1. ElixirDays (May) 2. Elixir Denver Meetup (June, online) 3. ElixirConf (Aug) 4. GambiConf (Nov) 5. Code BEAM Lite NYC (Nov) 6. @mozilla.ai Builders Demo Day (Dec) 7. SF Elixir Meetup (Dec)

090

Paulo Valente @polvalente.social · 08/12/2024

There's Integer.undigits to pair with list concatenation and Integer.digits!

1100

Paulo Valente @polvalente.social · 07/12/2024

Why would the Wizard of Oz help tho?

030

Paulo Valente @polvalente.social · 05/12/2024

I'd love to be added! Didn't find myself in there :)

020

Paulo Valente @polvalente.social · 03/12/2024

It would be extremely funny if someone used NxIREE to compile Nx to WASM and run it on this

050

Paulo Valente @polvalente.social · 30/11/2024

Wow they must be getting a lot of orders. Moon landers in the 70s only took ~3 days to ship, and now it's 2 weeks!

000

Paulo Valente @polvalente.social · 29/11/2024

Re: Gestures. I use the logitech mx vertical on my dominant hand, and a magic trackpad for scroll and other gestures on my nom-dominant hand. This divides a bit of the workload between both hands, and gets rid of the non-ergonomic scroll movement on the mouse hand.

110

Paulo Valente @polvalente.social · 28/11/2024

This also makes use of MLX's lazy evaluation to provide JIT compilation of Nx functions. The README says this, but just set it as your Nx backend and compiler, and it just works!

073

Reposted by Paulo Valente

Samrat Man Singh @samrat.me · 28/11/2024

The MLX(Apple Silicon) backend for Nx is now public on Github: #ElixirLang

14714

Paulo Valente @polvalente.social · 25/11/2024

Unified memory might make it so that certain combos of cpu+gpu in a CUDA+x86 environment become slower than the Mac counterpart just because of the transfer overhead! There's a lot of nuance to explore here, including some already-mapped (issues in their repo) inefficiencies in MLX itself

020

Paulo Valente @polvalente.social · 25/11/2024

The basic BERT example runs on 0.02~0.03s on EMLX with the GPU, and 0.03~0.04s on both EMLX and EXLA cpu. If you compare the averages of the ranges and want to have an nice number to show off, that's a 30% decrease in duration.

120

Paulo Valente @polvalente.social · 25/11/2024

The MLX (Apple Metal) backend for Nx!

060

Paulo Valente @polvalente.social · 25/11/2024

The homepage example from Bumblebee on using BERT is now running on EMLX! We should hopefully be releasing the library in the next few days. We're just missing a really small set of Nx.Backend calbacks and some housekeeping stuff

32010

Paulo Valente @polvalente.social · 24/11/2024

Ok, so I took this Axon example (hexdocs.pm/axon/your_fi...) for a spin on EMLX! A bit slower than Nx.BinaryBackend (runs the loop in 5s), but this is a simple model which really doesn't take advantage of a native platform. Next target is to get full Nx API coverage so we can run Bumblebee on it.

061

Paulo Valente @polvalente.social · 23/11/2024

To be fair, they're all operating on small tensors. Mostly for asserting correctness of the operations over input/option combinations

130

Paulo Valente @polvalente.social · 23/11/2024

That's funny... I could swear we started this endeavor 3 days ago, not 3 weeks! Jokes aside, big shout out to @samrat.me and @uwucocoa.moe who are taking MLX coverage by stride. We're almost at a point where we can run an Axon model! Bonus: this will be plug and play with Sharding

1124

Paulo Valente @polvalente.social · 20/11/2024

I'll work on better documenting that workflow, but you can take a look at github.com/polvalente/c... for an example

github.com

110

Paulo Valente @polvalente.social · 20/11/2024

Also, Nx can compile to wasm via NxIREE

140

Reposted by Paulo Valente

Mark Ericksen @brainlid.bsky.social · 19/11/2024

I'm looking for a remote, fulltime Elixir position! I've got lots of experience with Elixir, Phoenix, LiveView, mentoring, and AI. I'd love to work with you! Please reach out! #ai #llm #ElixirLang #MyElixirStatus #work #LiveView #ElixirPhoenix

24326

Paulo Valente @polvalente.social · 18/11/2024

This is picking up speed rather quickly, and we haven't even split workload yet!

070

Reposted by Paulo Valente

Samrat Man Singh @samrat.me · 18/11/2024

Making some progress towards an MLX backend for Nx #ElixirLang

1184

Paulo Valente @polvalente.social · 17/11/2024

Torchx is likely the closer and easier one! I suggest looking at how Torchx, via Torchx.Backend, implements add and dot. @cristine.bsky.social has also pinged me, so we can probably coordinate a larger effort on this. Let me know if you're on the Elixir or EEF Slack and we can coordinate there!

100

Paulo Valente @polvalente.social · 17/11/2024

It gets even worse as l and I look the same hahahaha!

100

Paulo Valente @polvalente.social · 17/11/2024

We have some good examples on how to bind C++ libs, and the abstraction is really similar to what we have. Should be easy to get something going quickly, and I would help with reviews and technical discussions throughout as well.

010

Paulo Valente @polvalente.social · 17/11/2024

Anyone want to help out in developing mlx bindings for Nx? #ElixirLang

6179

Paulo Valente @polvalente.social · 15/11/2024

I had a blast working on the project for this talk! Deploy Nx to the browser via WASM+LiveView is wild! #ElixirLang

1155

Paulo Valente @polvalente.social · 13/11/2024

I am! Flying from Rio, to land on Thursday morning!

020

Paulo Valente @polvalente.social · 12/11/2024

Yes, exactly! :erpc specifically makes it really easy to write code that assesses a single node and then just broadcast info to the whole cluster. And don't even get me started on process messaging and supervision.

000

Paulo Valente @polvalente.social · 11/11/2024

Last week we published the first installment on Nx Sharding on the @dockyard.bsky.social blog! dockyard.com/blog/2024/11... This feature set is really promising for the future of distributed #AI in #ElixirLang, both for training and inference.

dockyard.com

Nx Sharding Update - Part 1

This blog post is the start of a series in which I'll be providing updates on the development of data sharding in Nx.

061

Reposted by Paulo Valente

Mark Ericksen @brainlid.bsky.social · 05/11/2024

I'm open for work! I'd love to help #Elixir companies leverage #LLMs and #AI! I'm the creator and maintainer of the leading library github.com/brainlid/lan.... and host of the Thinking Elixir Podcast. I'm open to part-time and full-time positions. Please repost for reach! #MyElixirStatus

github.com

GitHub - brainlid/langchain: Elixir implementation of an AI focused LangChain style framework.

Elixir implementation of an AI focused LangChain style framework. - brainlid/langchain

23124

Reposted by Paulo Valente

polvo @coprodu.to · 02/11/2024

I am incredibly bullish on Elixir's data ecosystem - libraries like Nx, Livebook, and Explorer are evolving at an amazing pace. Had the opportunity to talk a bit to @polvalente.bsky.social today at @gambiconf.dev and was very impressed by his work on Nx :)

1131

Paulo Valente @polvalente.social · 31/10/2024

You have to use math to find the value

030