INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning

TL;DR

INTELLECT-2 is a pioneering 32-billion-parameter reasoning model trained via a fully decentralized, asynchronous reinforcement learning approach across a heterogeneous swarm of contributors, introducing new infrastructure and training techniques.

Contribution

It presents the first large-scale decentralized RL training framework for language models, with novel components and modifications that improve training stability and model performance.

Findings

Achieved state-of-the-art reasoning performance in 32B models.

Developed and open-sourced a complete decentralized training infrastructure.

Demonstrated successful training of a large-scale reasoning model without centralized control.

Abstract

We introduce INTELLECT-2, the first globally distributed reinforcement learning (RL) training run of a 32 billion parameter language model. Unlike traditional centralized training efforts, INTELLECT-2 trains a reasoning model using fully asynchronous RL across a dynamic, heterogeneous swarm of permissionless compute contributors. To enable a training run with this unique infrastructure, we built various components from scratch: we introduce PRIME-RL, our training framework purpose-built for distributed asynchronous reinforcement learning, based on top of novel components such as TOPLOC, which verifies rollouts from untrusted inference workers, and SHARDCAST, which efficiently broadcasts policy weights from training nodes to inference workers. Beyond infrastructure components, we propose modifications to the standard GRPO training recipe and data filtering techniques that were crucial to achieve training stability and ensure that our model successfully learned its training objective, thus improving upon QwQ-32B, the state of the art reasoning model in the 32B parameter range. We open-source INTELLECT-2 along with all of our code and data, hoping to encourage and enable more open research in the field of decentralized training.