Check out our new work on autonomous driving in new cities with map data + MARL!
Introducing 🥚EGGROLL 🥚(Evolution Guided General Optimization via Low-rank Learning)! 🚀 Scaling backprop-free Evolution Strategies (ES) for billion-parameter models at large population sizes
⚡100x Training Throughput
🎯Fast Convergence
🔢Pure Int8 Pretraining of RNN LLMs
We use EGGROLL 🥚to train RNN language models from scratch using only integer datatypes (and no activation functions!), scaling population size from 64 to 262144
2 (🐔🐔) orders of magnitude larger than prior ES works❗
EGGROLL 🥚for RL 🎮🤖
🥚 is competitive with, and in many cases, better than OpenES performance, even before considering the vast speed-up!
🥚 matched OpenES on 7/16 environments and outperformed it on another 7/16
🥚's low-rank approach does not compromise ES performance
Scaling LLM Reasoning with EGGROLL 🥚🧠📝
Using 🥚 to finetune RWKV-7 language models outperforms GRPO on Countdown and GSM8K ❗
🥚significantly outperformed GRPO on the Countdown task, achieving a 35% validation accuracy compared to GRPO's 23%❗
The EGGROLL Recipe
🧠🛠️ We replace full-rank perturbations with low-rank ones. Each update is still high rank, maintaining expressivity with faster training
🥚 EGGROLL converges to the full-rank update at a fast rate of 1/rank. The method is effective even with a rank of 1
Evolve at the hyperscale!
Work co-led with Mattie Fellows and Juan Agustin Duque.
Made possible by #Isambard and AIRR
🌐 Website: eshyperscale.github.io
📝 Paper: alphaxiv.org/abs/2511.16652
💻 Code: github.com/ESHyperscale...
🥚NanoEgg : github.com/ESHyperscale... (train in int 😉)
🥚EGGROLLing in the Deep with🚀 💯✕ Speedup
🥚 speed nearly reaches the throughput of pure batch inference, leaving OpenES far behind
🥚 reaches 91% of pure batch inference speed vs. OpenES reaching only 0.41%
General ML Training Made as Fast and Easy as Inference
1/ 🚗 🌏 What if an autonomous vehicle could move to a new city without collecting a single human demonstration in that city?
I am so excited to introduce our new work: Learning to Drive in New Cities Without Human Demonstrations.
