Prime Intellect releases algorithms layer for prime-rl codebase, shipping with GRPO, MaxRL, and four other pre-built methods
The update lets developers integrate custom algorithms via a single file
The update lets developers integrate custom algorithms via a single file
Commentary on X
Highest ranked@eliebakouch this looks like a massive step forward honestly.
View all
ECHO now officially part of prime-rl's Algorithms class. Super exciting to see more people play with it! I really believe that there's a ton to discover by learning from the terminal and designing input tasks with the goal of optimizing for self improvement without verifiers. https://twitter.com/PrimeIntellect/status/2074212134452633882