Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
This addresses the computational efficiency challenge in AI reasoning systems by enabling deeper computation without increasing token output or requiring specialized training data.
The authors tackled the problem of scaling test-time computation in language models by developing a novel architecture that performs latent reasoning through recurrent depth expansion, achieving performance improvements on reasoning benchmarks equivalent to models with up to 50 billion parameters.
We study a novel language model architecture that is capable of scaling test-time computation by implicitly reasoning in latent space. Our model works by iterating a recurrent block, thereby unrolling to arbitrary depth at test-time. This stands in contrast to mainstream reasoning models that scale up compute by producing more tokens. Unlike approaches based on chain-of-thought, our approach does not require any specialized training data, can work with small context windows, and can capture types of reasoning that are not easily represented in words. We scale a proof-of-concept model to 3.5 billion parameters and 800 billion tokens. We show that the resulting model can improve its performance on reasoning benchmarks, sometimes dramatically, up to a computation load equivalent to 50 billion parameters.