The fact that all LLM libraries don't have the same data format is as surprising as the fact that there is more than one sign language dialect

Ray is an excellent way of testing if all your `__repr__` are coded properly (but it shouldn't be)

Just stumbled upon RouteRL: a multiagent RL framework to facilitate the testing and development of efficient route choice strategies coexistence-project.github.io/RouteRL/ Looks pretty cool!

What is GGUF, Safetensors, PyTorch, ONNX? In this blog post, let's discover common formats for storing an AI model. huggingface.co/blog/ngxson/...

Today we're opensourcing MLGym, an API for AI research agents. MLGym relies on a gym environment that wraps a docker image. Each env has a task specified as a YAML file, telling in plain english what you want your LLM to achieve 👇

A few tips I share when I talk about perf with PyTorch in eager mode (with focus on small models): 🪢

I stand by my point that the word "agent" should be avoided at all costs. At least in RL, anytime I see an "Agent" class it's meant to be a "whatever doesn't fit in any other bucket in my codebase".

Everyone's like "hey I just coded and trained a SOTA LLM in my garage last week, also wrote a blogpost about it and opensourced the repo" and the only thing I did in the meantime was fix a CI and configure a remote interpreter on a server... 😢

A new release of tensordict is out github.com/pytorch/tens... Thanks to all who have contributed!

Clap hands if you were doing RL before it was cool

You’ve never really understood PyTorch until you’ve figured out what torch.scatter exactly does. I’ve never really understood PyTorch.

Introducing playground.mujoco.org Combining MuJoCo’s rich and thriving ecosystem, massively parallel GPU-accelerated simulation, and real-world results across a diverse range of robot platforms: quadrupeds, humanoids, dexterous hands, and arms. Get started today: pip install playground

One of the bests Mindscape episodes I've heard! I've always been interested in the epistemology of controversial scientific concepts such as emergence or consciousness. Vaguely defined concepts (or diverging definitions) are often at the root of many pointless debates. Kudos on clarifying things!

New Year’s resolutions: - eat healthier - exercise more - no more “is all you need” papers

Yesterday the hyped Genesis simulator released. But it's up to 10x slower than existing GPU sims, not 10-80x faster or 430,000x faster than realtime since they benchmark mostly static environments blog post with corrected open source benchmarks & details: stoneztao.substack.com/p/the-new-hy...

Wrong answers only: What does this `Human-computer` sticker seen at neurips hide?

Check out Motivo, a behavioral foundation model for humanoid control by FAIR. It's a one-of-its-kind unsupervised RL project, and it comes with a demo that is SO fun to play with! metamotivo.metademolab.com (for the record, they use compile and cudagraphs -> github.com/facebookrese...)

I’m 100% sure this button never does anything

PyTorch has released torchcodec yesterday, a powerful video decoding toolbox pytorch.org/blog/torchco... github.com/pytorch/torc...

Tomorrow with Matteo Bettini we'll be presenting BenchMARL at #NeurIPS (@neuripsconf.bsky.social) in #Vancouver

I'm excited to be in #Vancouver for #NeurIPS 2024 where I brought a bunch of @LEGO_Group bricks in my bag! *no, I'm not Santa!

We’re looking for an intern (research scientist/PhD) to join the PyTorch team in NYC this summer and work on GPU kernel generation, more info here: www.metacareers.com/jobs/6044463... You’ll be working with @soumithchintala.bsky.social, @marksaroufim.bsky.social and myself

There are two kinds of developers: - Those who know what they would have done differently had they known what a pain it was going to be - Those who change things accordingly and ruin everyone's life forever

My turn to go on the MLOps podcast! We talked about pytorch, compile, RL and lots of other stuff. Check it out wherever you listen to podcast, as they say 🍏 podcasts.apple.com/gb/podcast/m... ᯤ open.spotify.com/show/7wZygk3...

define "amusingly"

bsky pro-tip: To check what lists you're part of: clearsky.app/%3Cid%3E.bsk... You're welcome!

👉 PyTorch starter pack! go.bsky.app/P3bo7mp

You know this annoying thing where you're thinking "I'd like to see the code of this pytorch class!" and you cick on source and it leads you to some weird looking html page? Fear no more! Now (on the `main` doc) you'll end up right on github!

If you'd like to influence what features the PyTorch distributed team work on in torchtitan (e.g. MoE, multimodal, context parallelism, etc.), go made your voices heard here!

PyTorch internals by @ezyang.bsky.social This post is a long form essay version of a talk about PyTorch internals, that he gave at the PyTorch NYC meetup on May 14, 2019. blog.ezyang.com/2019/05/pyto...

Your worst fear in three words? OpenReview Data Leak

Hello BlueSky! Joao Henriques (joao.science) and I are hiring a fully funded PhD student (UK/international) for the FAIR-Oxford program. The student will spend 50% of their time @UniofOxford and 50% @MetaAI (FAIR) in London, while completing a DPhil (Oxford PhD). Deadline: 2nd of Dec AOE!!

For those who wonder about the best way to start contributing to pytorch or open-source projects, here are the top three pointers I'd share: 1. The Ultimate Guide to PyTorch Contributions github.com/pytorch/pyto... For pytorch core that should be the n1 item on your list.

RLC will be held at the Univ. of Alberta, Edmonton, in 2025. I'm happy to say that we now have the conference's website out: rl-conference.cc/index.html Looking forward to seeing you all there! @rl-conference.bsky.social #reinforcementlearning

I have a deep, dark secret ... I write most of my docstrings using LLMs: I copy paste the function and ask for the docstrings. Minor edits and you're done (usually I write the example myself, that's why they're so bad). Most of the time it's *much* better than whatever I would have done!

One of my fav projects: LeanRL, a simple RL library that provides recipes for fast RL training using torch.compile and cudagraphs. Using these, we got >6x speed-ups compared to the original CleanRL implementations. github.com/pytorch-labs...

Not sure this makes it "10x better" than anything else, but I wrote a tuto on how to export trained policies in torchrl to use them on any edge device or hardware (from RaspberryPi to your smart toaster) The tuto is short bc it's easy Feedback is welcome!! pytorch.org/rl/main/tuto...

Just got started with the PyTorch Starter Pack Who did I forget?

Stable-Baselines3 v2.4.0 has been released! (along with SB3-Contrib and RL Zoo) We added a new algorithm in SB3-Contrib, CrossQ, a sample-efficient alternative to SAC, and added support for Gymnasium v1.0. Full changelog: github.com/DLR-RM/stabl...

That’s so true! There’s some stigma associated with RL and I see a bunch of people doing everything they can to avoid putting that label on what they do, even if they have MDPs, rewards, value fn and all the rest. There’s so much domain knowledge being lost, and so much friction for collaborations.

The RL (and some non-RL folks) starter pack is almost full. Pretty clear that the academic move here has succeeded go.bsky.app/3WPHcHg

Hello bluesky! Happy to be here and not there I'm a research engineer at #Meta / #PyTorch (order tbd). I work on torchrl github.com/pytorch/rl (sort of a swiss knife for RL in PyTorch) but what I'm secretly most proud of is tensordict github.com/pytorch/tens... - I can be pretty annoying about it.