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
👇

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

Built on TorchRL and PyTorch, BenchMARL ensures high performance and state-of-the-art implementations, while its flexible configuration and standardized reporting make it a breeze to use.

BenchMARL is a cutting-edge training library designed to bring standardized benchmarking to the world of Multi-Agent Reinforcement Learning (MARL). It allows for easy comparison across different algorithms, models, and environments, making it a game-changer for researchers and developers alike.

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

We believe @PyTorch should have a stronger place in STEM education. With BricksRL, we aim to provide a fun and interactive way to learn about control and Reinforcement Learning, inspiring the next generation of researchers and engineers

Why Lego? It offers low-cost hardware for learning & experimenting with control or Reinforcement Learning in the real world. Standardized, reproducible, and fun! Plus, the Lego community is active and vibrant, making it a great fit for our project.

I'm teaming up with Sebastian Ditter & @gdefabritiis.bsky.social to present our paper on BricksRL, a library that enables control of Lego robots using #PyTorch

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!

They've already taken over london!

define "amusingly"

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!

Not only that, but you also get a better utilization of your GPU: the total usage with compile and cudagraphs is lower than eager, compile or cudagraphs in isolation, meaning that you could for instance easily train two networks at the same time.

We get an 6.8x speed-up with PPO (Atari), 5.7x for SAC, 3.4x for TD3 and 2.7x for PPO with continuous actions.

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...

I stand by my point that any interesting math equation is better understood in code format