Podcast: ThinkingElixir 130 - Bumblebee Buzz with José Valim

Episode 130 of Thinking Elixir. The Elixir community is buzzing with excitement around Bumblebee! Bumblebee makes it easy to use a variety of pre-trained neural network models in Elixir. José Valim, Paulo Valente, and Jonatan Kłosko join us to talk about this latest advancement. We learn how Bumblebee works, the abilities that are supported today, where things are going next and the longer term visions of what’s possible. We cover the new Livebook smart cells that create an easy playground for playing with Bumblebee, a single-file Phoenix example that does image classification, and much more!


I think I’m going to have to listen to this a couple of times to distill all the info about ML+Elixir :lol:

The first 5 minutes has some really good info about some of the AI/ML stuff in the news right now too, specifically, how ChatGPT and Dalli aren’t open source even though they are from a company called ‘OpenAI’!!


Ha! I know! Ironic isn’t it. I found it Jose’s statement interesting about complex/advanced models potentially becoming so large that individuals couldn’t create, run or test them on the hardware we have.

It’s an interesting time for ML!


I wonder if someone has started a petition asking them to change it, or reported them to some standards authority as misleading… because it definitely is :lol: When I first visited their website my immediate thought was that it sounds cool because it appears to be open source. If you look at their charter it follows the same theme; they are trying very hard to come across as an ‘open’ (/open source) organisation. It makes me super suspicious :105:

It is! And I was curious as well after I reached that part of the podcast… so googled ‘best GPU for deep learning’, and it seems to be…


It has 80GB of RAM!

NVIDIA A100 is the world’s most advanced deep learning accelerator. It delivers the performance and flexibility you need to build intelligent machines that can see, hear, speak, and understand your world. Powered by the latest NVIDIA Ampere architecture, the A100 delivers up to 5x more training performance than previous-generation GPUs.

Not cheap at $13,999 :lol:

How much for a full spec’d system?

$187K :043:

(Disclaimer: nobody buy that system! I just added things to get an idea of cost, some of those items may not be necessary/ideal)