Masafumi Enomoto work at NEC Corporation.

Posts

DISC: Dynamic Decomposition Improves LLM Inference Scaling

/in /by

Inference scaling methods for LLMs often rely on decomposing problems into steps (or groups of tokens), followed by sampling and selecting the best next steps. However, these steps and their sizes are often predetermined or manually designed based on domain knowledge. We propose dynamic decomposition, a method that adaptively and automatically partitions solution and reasoning traces into manageable steps during inference. By more effectively allocating compute — particularly through subdividing challenging steps and prioritizing their sampling — dynamic decomposition significantly improves inference efficiency. Experiments on benchmarks such as APPS, MATH, and LiveCodeBench demonstrate that dynamic decomposition outperforms static approaches, including token-level, sentence-level, and single-step decompositions, reducing the pass@10 error rate by 5.0%, 6.7%, and 10.5% respectively. These findings highlight the potential of dynamic decomposition to improve a wide range of inference scaling techniques.

Read more

On Synthesizing Data for Context Attribution in Question Answering

/in /by

Question Answering (QA) accounts for a significantportion of LLM usage “in the wild”.However, LLMs sometimes produce false ormisleading responses, also known as hallucinations.Therefore, grounding the generatedanswers in contextually provided information—i.e., providing evidence for the generated text—is paramount for LLMs’ trustworthiness. Providingthis information is the task of context attribution.In this paper, we systematically studyLLM-based approaches for this task, namelywe investigate (i) zero-shot inference, (ii) LLMensembling, and (iii) fine-tuning of small LMson synthetic data generated by larger LLMs.Our key contribution is SYNQA: a novel generativestrategy for synthesizing context attributiondata. Given selected context sentences, anLLM generates QA pairs that are supported bythese sentences. This leverages LLMs’ naturalstrengths in text generation while ensuring clearattribution paths in the synthetic training data.We show that the attribution data synthesizedvia SYNQA is highly effective for fine-tuningsmall LMs for context attribution in differentQA tasks and domains. Finally, with a userstudy, we validate the usefulness of small, efficientLMs (fine-tuned on synthetic data fromSYNQA) in context attribution for QA.

Read more

DISC: Dynamic Decomposition Improves LLM Inference Scaling (SSI-FM)

/in /by

Inference scaling methods often rely on decomposing problems into steps, followed by sampling and selecting the best next steps. However, these steps and their sizes are typically fixed or depend on domain knowledge. We propose dynamic decomposition, a method that adaptively and automatically breaks down solution and reasoning traces into manageable steps during inference. By allocating compute more effectively, particularly by subdividing challenging steps and sampling them more frequently, dynamic decomposition significantly enhances inference efficiency. Experiments on benchmarks such as APPS, MATH, and LiveCodeBench demonstrate that dynamic decomposition outperforms static approaches, including token-level, sentence-level, and single-step decompositions. These findings highlight the potential of dynamic decomposition to improve a wide range of inference scaling techniques.

Read more

DISC: Dynamic Decomposition Improves LLM Inference Scaling (DL4C)

/in /by

Inference scaling methods often rely on decomposing problems into steps, followed by sampling and selecting the best next steps. However, these steps and their sizes are typically fixed or depend on domain knowledge. We propose dynamic decomposition, a method that adaptively and automatically breaks down solution and reasoning traces into manageable steps during inference. By allocating compute more effectively—particularly by subdividing challenging steps and sampling them more frequently—dynamic decomposition significantly enhances inference efficiency. Experiments on benchmarks such as APPS, MATH, and LiveCodeBench demonstrate that dynamic decomposition outperforms static approaches, including token-level, sentence-level, and single-step decompositions. These findings highlight the potential of dynamic decomposition to improve a wide range of inference scaling techniques.

Read more