text stringlengths 0 2.02k |
|---|
**Adam Stacoviak:** Yeah. Well, that speaks to the maturity we talked about earlier though, right? Like when you said, how has open source matured over the years? I think that speaks to the maturity of it, because Brett’s perspective from what I recall in the post is his contribution to Python itself. Whereas your all ... |
\[50:22\] So it kind of depends... I mean, licenses play a role in helping that line and being crossed too, because what makes open source open source is, one, I would suppose OSI’s blessing of the license... But then, two, its permissiveness, and its restrictions. |
**Jerod Santo:** Right. And AGPL is not one of the more permissive licenses, but we don't wanna necessarily bike-shed licenses at the moment. I will say that -- that’s why I started off with it’s so important to set expectations. So I think you can layer on top of this gift concept and hold yourself to a higher standar... |
I remember, we just had Richard Hipp back on the show, from SQLite. Another one of our greatest hits, episode 201; check it out. He’s one of my favorite guests. He says they want to maintain SQLite until 2050. That’s something that he holds himself to, and he puts that out there. Now can they get that done or not? Not ... |
But the beauty of open source if GitLab suddenly lets all of us down, or Forem, God forbid, goes against their previous statements - like, fork it and start a revolution. That’s the freedom in open source. Spell it backwards. Or Moref, I don't know. Start a Moref project. \[laughter\] |
**Adam Stacoviak:** Right. Like Deno and Node. |
**Jerod Santo:** That's right. "Not less F, we need more F around here." \[laughter\] Sorry. But that's the cool thing about open source; that was a gift, as of that point. Now, maybe the next point it moves to something that you don't want anymore; that sucks, and maybe you are let down by those people, but there are ... |
**Adam Stacoviak:** Yeah, precisely. It's copyrighted, it's intellectual in terms of IP... It may not be intellectual in terms of its actual codebase... Whatever. \[laughter\] |
**Jerod Santo:** Right. |
**Adam Stacoviak:** Yeah, for sure. And that's the beauty of open source - we can all show up and accept this free gift, and we have choices based upon that. Whereas if its proprietary, there is no gift and there is no option, really. |
**Jerod Santo:** Which is why I'm an advocate for people very clearly communicating on their own projects - the projects that you create and run, very clearly communicate the expectations of you and of the community... One of the questions we ask many people - what kind of open source project is this? Is this open sour... |
And then there are more things people should expect than just what we've talked about with Brett on that episode, but... I think that's a baseline. |
**Christina Gorton:** I definitely have a backlog of podcasts to listen to now, so I really appreciate talking to you all. |
**Adam Stacoviak:** You're welcome. |
**Christina Gorton:** Adam, Jerod - thank you so much for joining us today. We've really enjoyed it. |
**Adam Stacoviak:** It was awesome. Thanks for having us. |
**Jerod Santo:** Yeah, thanks for having us. It's been lots of fun. |
• José Valim introduces Nx, a library for numerical computing, machine learning, and data science in Elixir |
• Nx includes multi-dimensional tensors and numerical definitions for efficient GPU computation |
• José Valim discusses the inspirations for Nx, including Jax and Google XLA |
• Nx has bindings for EXLA (Google XLA) and is working on bindings for PyTorch (LibTorch) |
• Other libraries released by Nx include Axon (high-level neural network library) and LiveBook (interactive and collaborative code notebooks) |
• José Valim explains the motivation for developing Nx, driven by the desire to expand the capabilities of Elixir and make it a more diverse and powerful language |
• Expanding Elixir's capabilities to include data processing and machine learning |
• Bringing machine learning to Elixir to provide a more comprehensive toolset |
• José Valim's interest in broadening Elixir's domains and areas of use |
• Collaboration with Sean Moriarity on developing Elixir-based machine learning capabilities |
• Comparison with existing Python libraries and tooling for machine learning and data science |
• Goal of making Elixir a more viable option for developers working in machine learning and data science |
• Discussion of the potential for Elixir to replace Python in certain areas of AI and machine learning development |
• The speaker and Sean discovered the Jax library and its functional programming style, which led to the development of Nx, a numerical computation library for Elixir. |
• Jax's computation graph approach is discussed, where the library builds a graph of operations and then compiles it to run on the GPU. |
• The speaker notes that immutability in Elixir is a key feature that allows Nx to avoid some pitfalls present in Jax, such as the tape pattern. |
• Axon, a neural network library built on top of Nx, is introduced, which allows users to build and train neural networks using a high-level API. |
• The speaker expresses a desire for a GPU to run some examples in Axon, but is currently hindered by supply chain issues and NVIDIA's recent changes to their GPU lineup. |
• Introduction to Axon, a high-level API for building machine learning models in Elixir |
• API design and usability, including interoperability with other frameworks (e.g. PyTorch, TensorFlow) |
• Serialization of models to multiple formats, including ONNX |
• Potential use cases for Axon in edge computing and embedded systems |
• Collaboration and community involvement in Axon development |
• History and development timeline of Axon and related projects (Nx, XLA) |
• Productivity and efficiency in building machine learning libraries, comparing Axon's approach to others in the field |
• Axon's API is designed to be familiar and build upon existing knowledge |
• Inspiration from other projects, such as Think AI in Python and PyLightning |
• Diversity of AI models and frameworks, with consistent architecture patterns |
• Axon's design goal of being easy to understand for developers outside of the Elixir community |
• Development of LiveBook, an interactive and collaborative notebook for Elixir |
• LiveBook's features and goals, including real-time collaboration and interactive data inspection |
• Plans to expand LiveBook's capabilities to include interactive data analysis and neural network training |
• Problems with Jupyter notebooks, including formatting, version control, and dependencies |
• LiveBook, a new approach to collaborative notebooks with features like explicit dependencies and reproducibility |
• Inspiration from other tools, including LiveMarkdown, Jupyter Notebooks, Pluto.jl, and Deepnote |
• LiveBook's use of LiveView and Monaco editor for autocompletion and collaborative features |
• Distributed notebooks and ability to run on multiple machines |
• Plans for future development, including graphs and data frames |
• Challenges of supporting notebooks in production environments |
• Erlang Ecosystem Foundation Machine Learning Working Group |
• LiveBook features and limitations, including sequential evaluation and collaboration |
• Integration with tools like TensorBoard and potential for unified monitoring of training runs |
• Collaboration features, including GitHub integration and pluggable file systems |
• Nx and its role in bringing machine learning tools to Erlang and Elixir |
• Community and involvement opportunities, including Discourse forum, Slack, and monthly meetings |
• José Valim's work on building a neural network in Erlang using Axon and Nx |
• Possibility of cross-over from the Python world to Erlang for AI development |
• Plans for Daniel Whitenack to try out and share José's work with the Practical AI community |
• Links to José's LiveBook demo and the Erlang Ecosystem Foundation in the show notes |
**Jerod Santo:** Alright, I'm joined by José Valim, creator of Elixir and frequent guest on the Changelog. I think this is your fourth time on the show. Welcome back. |
**José Valim:** Thank you. Thanks for having me again. |
**Jerod Santo:** Excited to have you. Lots of interesting stuff going on in your neck of the woods. And I'm also joined by -- hey, that's not Adam. That is Practical AI co-host, Daniel Whitenack. What's up? |
**Daniel Whitenack:** Practical AL. |
**Jerod Santo:** Yeah. |
**Daniel Whitenack:** Yeah... Practical AI sometimes, with the font, on Zoom, it looks like Practical AL... So when we record on our podcast, normally I'm known as Practical AL. |
**Jerod Santo:** Well, welcome to the show. I have a Tool Time reference. You'll be my Al Bundy for this show. But that would be too old for most people to get that one. Or you can be my Adam, I'll be your Chris Benson, and we'll co-host this sucker, how about that? |
**Daniel Whitenack:** That sounds wonderful. I'm excited to be here. |
**Jerod Santo:** Well, I had to call in the big guns, because I know very little about this space. In fact, everything I know about the world of artificial intelligence, I learned from producing practical AI, and by listening to Practical AL do his thing each and every week... So that's why Daniel is here. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.