Chopthin method boosts diversity and accuracy in AI reasoning sampling
Chopthin-Consensus Power Sampling: A Diversity-Preserving Approach to LLM Decoding
Computation and LanguageArtificial Intelligence
Summary
Large language models often try to figure out the best answers by exploring many possible reasoning paths, but some methods quickly throw away less likely paths, which might still be good. The authors introduce a new way called Chopthin-Consensus Power Sampling that keeps a more balanced and diverse set of answers without forcing equal importance on all paths. They also group similar answers to pick the most popular one, improving the final result. Tests show this method helps AI models find better answers more often without retraining.
What this means in practice
- •For machine learning engineers: Improve reasoning accuracy in language model applications by preserving diverse inference paths during token sampling.
- •For natural language processing developers: Develop language generation tools that select better final answers by merging semantically similar outputs from diverse decoding trajectories.
Authors
Minoo Ahmadi, Seyedarmin Azizi, Erfan Baghaei Potraghloo, Mehdi Kamal, Massoud Pedram
Abstract
Inference-time power sampling via Sequential Monte Carlo (SMC) can substantially improve large language model (LLM) reasoning without requiring post-training. However, many existing SMC approaches rely on equal-weight resampling, which can aggressively prune low-weight trajectories, discarding potentially correct reasoning paths and degrading the genealogical diversity of the search space. To address this, we introduce Chopthin-Consensus Power Sampling (CCPS). Our method applies the Chopthin resampler to LLM decoding: rather than equalizing weights and forcing unnecessary particle duplication, it enforces an upper bound on the ratio between the largest and smallest weights and carries the unequal weights forward. This targeted intervention preserves a richer set of distinct reasoning paths, keeps the weighted SMC approximation unchanged in conditional expectation, and guarantees a lower bound on the post-resampling effective sample size (ESS). To fully exploit this enriched population, we employ a semantic-majority selection mechanism that merges token-identical final trajectories, clusters semantically equivalent answers, and returns the answer supported by the largest number of distinct trajectories. Evaluating across three open-weight models and five reasoning benchmarks, we show that Chopthin increases oracle coverage in 13 of 15 settings. Combined with semantic-majority selection, CCPS matches or exceeds the final-answer accuracy of the Power-SMC baseline in 14 of 15 settings, delivering absolute gains of up to 10.6 percentage points. These findings demonstrate that diversity-preserving resampling and diversity-aware selection are complementary mechanisms for training-free LLM reasoning. Code is available at github.com/MinooAhmadii/chopthin-consensus-power-sampling.