Search papers, labs, and topics across Lattice.
This paper introduces NF-CoT, a latent reasoning framework that leverages normalizing flows to enhance the reasoning capabilities of large language models by allowing intermediate computations to occur in continuous states rather than through discrete text. By integrating a TARFlow-style normalizing flow within the LLM backbone, NF-CoT maintains the advantages of chain-of-thought reasoning while enabling efficient probabilistic sampling and left-to-right generation. The results show that NF-CoT significantly improves code-generation pass rates compared to explicit chain-of-thought and previous latent reasoning methods, while also reducing the computational cost of intermediate reasoning steps.
NF-CoT achieves higher code-generation accuracy with lower reasoning costs by enabling latent reasoning in continuous states without sacrificing the benefits of traditional chain-of-thought approaches.
Large language models often improve reasoning by generating explicit chain-of-thought (CoT), demonstrating the importance of intermediate computation. However, textual CoT forces this computation through a discrete, serial, and communication-oriented token stream: each reasoning step must be verbalized before the model can proceed, even when the underlying update is semantic, uncertain, or only partially formed. Latent reasoning offers a higher-bandwidth alternative by performing intermediate computation in compact continuous states before committing to text. Yet existing latent-reasoning methods often sacrifice key advantages that make CoT effective in autoregressive language models, including native left-to-right generation, probabilistic sampling, compatibility with KV-cache decoding, and tractable likelihood estimation. We propose NF-CoT, a latent reasoning framework that preserves these advantages by modeling continuous thoughts with normalizing flows. NF-CoT instantiates a TARFlow-style normalizing flow inside the LLM backbone, defining a tractable probability model over compact continuous thoughts distilled from explicit CoT. Continuous-thought positions are generated by an NF head, while text positions are generated by the standard LM head within the same causal stream. This design provides exact likelihoods for latent thoughts, enables probabilistic left-to-right decoding with the original KV cache, and supports direct policy-gradient optimization in the latent reasoning space. On code-generation benchmarks, NF-CoT improves pass rates over explicit-CoT and prior latent-reasoning baselines while substantially reducing intermediate-reasoning cost.
