A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM
A*-Thought-V2 compresses redundant chain-of-thought steps into latent tokens guided by hidden-state geometry, improving accuracy up to 2.6% while halving response length.
A*-Thought-V2 models chain-of-thought as a hidden-state trajectory projected into a 3D PCA space and compresses steps whose transitions deviate from the question-to-solution direction into continuous latent tokens, keeping aligned steps explicit. Training uses stepwise embedding forcing and label forcing with soft multi-modal vocabulary supervision. On Qwen3.5-9B and Qwen3.6-27B across six in-domain and out-of-domain benchmarks it improves average accuracy by up to 2.6%, cuts response length by up to half, and raises Accuracy per Computation Unit 2.29x while reducing preprocessing and training time by 94.6% and up to 80.3%.
- Explicit-implicit interleaved latent architecture replaces hard step pruning in chain-of-thought
- Directional angles classify steps: small for direct execution, large for checking and exploration
- Up to 2.6% accuracy gain with half the response length on Qwen3.5-9B and Qwen3.6-27B
- 2.29x Accuracy per Computation Unit; 94.6% and up to 80.3% preprocessing/training time cuts
- Tags
- chain-of-thoughtefficiencyhidden-state-dynamicslatent-reasoningqwen3.5-9bqwen3.6-27btest-time-compute
- AI models
- A*-Thought-V2Qwen3.5-9BQwen3.6-27B
From the original article
1 min read at arxiv.orgChain-of-Thought (CoT) improves the reasoning ability of Large Language Models (LLMs) but incurs substantial computation and context costs. Existing methods either lose intermediate information through hard pruning or lack a principled criterion for continuous compression.
By Xiaoang Xu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.