Daniel Falbel

YouTube video by useR! Conference Keynote #2 R in the AI Era Leveraging Modern Technologies in Practice - Simon Urbanek

This was really nice to watch #rstats :) Fun application of attention models. Also nice to see torch used for such important work! youtu.be/0Zra1WYoiOM?...

With a nice latent space vis ;)

luz v0.5.1 is now on #rstats CRAN. Just a small bug fix related to forwarding `predict` parameters to the model. I also added a new Variational Autoencoder example to our examples gallery: mlverse.github.io/luz/articles...

Refer to kernel laucnhing thread instead of assuming the main thread by dfalbel · Pull Request #1455 · ipython/ipykernel This PR proposes a patch to address the issue described in #1438 (comment) With this, we no longer asssume that the IPykernel app starts from the main thread. Instead, we keep a reference to the la...

As an #rstats dev, I never expected writing a PR to ipykernel :) github.com/ipython/ipyk... It's now running for everyone with ipykernel >= 7.1

It has many limitations:

- no dynamic shapes
- hardcoded weights

But it should be possible to greatly improve this in the future.

Where `resnet18_stablehlo.mlir` is a model exported with:

Here's a small example:

Embedding Atlas by dfalbel · Pull Request #124 · tidyverse/ragnar Add support for visualizing the store using embedding-atlas

The motivation for it comes from working on inegrating
Apple embedding atlas into ragnar: github.com/tidyverse/ra...

We ended up using the python package, but we could have used a #rstats / JS only solution if we had some nice mosaic integration.

Just made a nice website for mosaicr. A small #rstats htmlwidget package that allows using
@idl.uw.edu Mosaic data visualization framework from R.

The real power comes when embedding Mosaic plots within Shiny apps. See examples and docs in

dfalbel.github.io/mosaicr/

r-xla r-xla has 6 repositories available. Follow their code on GitHub.

Yep! We have some work in github.com/r-xla
It's a long term project though. Theoretically it's already possible to create a graph in jax , export it to stablehlo and ecute it in R with no python dep.

minhub/R/gemma3.R at main · mlverse/minhub Minimal implementation of Deep Learning models. Contribute to mlverse/minhub development by creating an account on GitHub.

tiny ( <400loc) implementation of Gemma3 270M using torch for #rstats

github.com/mlverse/minh...

Includes support for loading weights from hugging face hub.

Release v0.2.0 · mlverse/tok Updated upstream tokenizers to 0.20.3 Update extendr-api to 0.8.1

tok is back on CRAN! tok is @hf.co tokenizers for #rstats. It uses the same Rust libray as the python interface. It's pretty fast and fully compatible with tokenizers.json available on hub.

See the release notes:

github.com/mlverse/tok/...

Dev 📦 alert! {lang} translates R help on-the-fly using your local LLM! It also overrides the `?` so you can easily access the translated docs and have them displayed on your IDE's help pane github.com/mlverse/lang #rstats #llm #ollama

R-Universe Named R Consortium Top-Level Project We're excited to announce R-Universe has been named the R-Consortium's newest Top-Level Project.

Excited and grateful that R-Universe is R Consortium's newest top-level project! This means sustained support for @rOpenSci.hachyderm.io.ap.brid.gy's platform for discovery and publishing of #rstats packages. Hats off to @jeroenooms.bsky.social for his leadership!

ropensci.org/blog/2024/12...

Happy to help making it work on MPS. It should just work if you create a tensor on the MPS device, using eg:

torch_randn(10,10, device="mps")

torch for #rstats has reached 500 GitHub stars :)

Changelog

📦 usethis 3.1.0 📦 is released. `use_vignette()` and `use_article()` can now help you initiate a Quarto (.qmd) vignette or article. #rstats

usethis.r-lib.org/news/index.h...

This is only happens on Linux when using the CPU. But many workflows involve transforming data on the CPU before executing the model on the GPU, so this fix should improve performance of most torch programs running on Linux!

But LibTorch is really optimized to work with Intel MKL, and indeed statically links to a version of MKL for faster math kernels, but we were not using it leaving a lot of performance on the table.

The problem was that previous versions of torch would use whatever BLAS library R is configured to use (which in most system will be a single threaded BLAS, and if you had some fun configuring R, it could be OpenBLAS.

This includes a bug fix that makes #rstats torch at least 2x faster on Linux CPU's, and possibly even more for other workflows.

Libtorch 2.5.1 by dfalbel · Pull Request #1204 · mlverse/torch Add support for LibTorch 2.5.1 TODO: Update Libtorch M1 builds - and patch the allocator Regenerate R level wrappers Figure out pre-built binaries for x86_64 macOS Drop unsupported CUDA from C...

Finally merged support for LibTorch 2.5 into #rstats torch main! github.com/mlverse/torc...

Nice! Wrapping the bucket with a CDN, such as CloudFront, should also make downloads faster, due to regional caching and other optimizations that CDN's do. It also allows you to decouple the storage location from how things are distributed, so it gives you freedom to change storage structures later.

I know nothing about nix, but how do folks download from the cache later? In general using a CloudFront distribution instead of directly accessing the bucket will reduce a lot the egress costs.

After long investigation, I still don't understand what is really happening and what's wrong, but... The fix turned out being quite simple. Just had to add the RTLD_DEEPBIND flag when opening the dynamic library 🤷‍♂️

github.com/mlverse/torc...

MKL and openBLAS interactions - a question about linking I'm using a binary (R) that dynamically links to a generic version of BLAS, for instance (and in a lot of cases) this is openBLAS. Now, inside R, I'm dynamically loading another shared library (lib...

Just posted a Stack Overflow question! First one in a couple years. Do people still use it? stackoverflow.com/q/79171784/3...

S7 is really nice! I've exploring it in the micrograd package, a minimal implementation of an automatic differentiation engine (think torch's autograd) built from scratch in R: github.com/dfalbel/micr...

YouTube video by Posit PBC Introducing Positron, a new data science IDE - posit conf 2024

Our #PositConf2024 talk(s) on Positron are now available on YouTube! It was really special to get to introduce this new project to folks ❤️ and I'm excited this is broadly available now.

Take a look if you are curious about our new IDE:
youtu.be/8uRcB34Hhsw