Latent Recurrent Thoughts for Frozen LLMs
Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs
LRT achieves better reasoning performance than chain-of-thought prompting while using significantly less compute, making it efficient for developers working with frozen models.
Latent Recurrent Thoughts (LRT) improves reasoning with frozen large language models by using continuous representation space instead of discrete token space.
The method combines a task-dedicated proposer with a recurrent reasoner that refines latents through bounded residual corrections.
LRT outperforms prior frozen-decoder methods on symbolic reasoning tasks like Countdown-4 and Sudoku, and natural-language reasoning benchmarks including HumanEval and MBPP.
Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs
Chain-of-thought reasoning unfolds in discrete token space: each step is committed as text, errors propagate, and eliciting good traces presupposes traces to imitate. Reasoning instead in a model's continuous representation space - where intermediate states are vectors rather than words - sidesteps these constraints, but leaves open how those latent states should be computed. We approach this along two axes. First, we keep a large language model (LLM) frozen and use it for what it is already good at - modeling and decoding sequences - while a small auxiliary network supplies continuous latent thoughts as input. Second, we produce those latents by recurrence: a tiny recurrent reasoner refines them over many steps, decoupling the depth of computation from the size of the model, so that the latents are a product of iterative processing rather than a single forward pass. We instantiate this as Latent Recurrent Thoughts (LRT): a task-dedicated proposer supplies base latents, a recurrent reasoner refines them through bounded residual corrections, and the frozen LLM decodes the answer. On symbolic reasoning with answer supervision but no reasoning traces (Countdown-4, Sudoku) and on natural-language reasoning (HumanEval, MBPP, StrategyQA), LRT substantially outperforms prior frozen-decoder continuous-space reasoning methods under an identical decoder, prompt, data, and training budget, and outperforms non-thinking-mode chain-of-thought prompting on the same backbone at a small fraction of its inference compute.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include LRT, reasoning models.
Source attribution: curated from arXiv cs.AI (@Zhaoliang Chen, Jie Fu).