RL @LumaLabsAI | cs @uwaterloo, prev reasoning @cohere
Joined May 2024
- Tweets28
- Following369
- Followers857
- Likes547
Spent some time creating a lightweight, performant async RL stack in pure JAX with @AdityaMakkar0! We share a work log covering inference, RDMA weight transfer, memory optimizations, sharding and a few more JAX engineering insights we've seen along the way! It’s a bit of a long read, but I hope it sheds some light on the systems side of scaling RL!
Divya Makkar retweeted
Excited to share some of the work we’ve been doing at @gen_intuition! Together with @kyutai_labs and in collaboration with Epic Games we introduce MIRA, a real-time multiplayer world model! If you’re at ICML 🇰🇷 feel free to drop by Booth 111 to chat more about MIRA and world-models!
Introducing MIRA.
A playable, multiplayer world model.
A dream of Rocket League.
Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press.
Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League.
▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira
At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.
I spent the past few months building JAXformer: One of the first open source guides on how to scale modern transformers in JAX.
Trained entirely on TPUs, it supports distributed ML, Ray tokenization, MoE, n-D parallelism and end-to-end inference.
Here’s how to do it:
We trained until we hit a 3.28 val loss (inspired by nanoGPT speedrun) over 26,100 steps with 6.5 billion tokens.
1
27
3,761
I worked on this with @AdityaMakkar000 and @chinmayjindal_. Take a look at our blog to see how you can go from zero-to-one scaling a transformer from scratch.
GitHub: github.com/divyamakkar0/JAXf…
Blog: jaxformer.com/
