Pay for Hints, Not Answers: LLM Shepherding for Cost-Efficient Inference

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dong, Ziming, Sharma, Hardik, O'Toole, Evan, Champati, Jaya Prakash, Wu, Kui
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914292073234432
author Dong, Ziming
Sharma, Hardik
O'Toole, Evan
Champati, Jaya Prakash
Wu, Kui
author_facet Dong, Ziming
Sharma, Hardik
O'Toole, Evan
Champati, Jaya Prakash
Wu, Kui
contents Large Language Models (LLMs) deliver state-of-the-art performance on complex reasoning tasks, but their inference costs limit deployment at scale. Small Language Models (SLMs) offer dramatic cost savings yet lag substantially in accuracy. Existing approaches - routing and cascading - treat the LLM as an all-or-nothing resource: either the query bypasses the LLM entirely, or the LLM generates a complete response at full cost. We introduce LLM Shepherding, a framework that requests only a short prefix (a hint) from the LLM and provides it to SLM. This simple mechanism is surprisingly effective for math and coding tasks: even hints comprising 10-30% of the full LLM response improve SLM accuracy significantly. Shepherding generalizes both routing and cascading, and it achieves lower cost under oracle decision-making. We develop a two-stage predictor that jointly determines whether a hint is needed and how many tokens to request. On the widely-used mathematical reasoning (GSM8K, CNK12) and code generation (HumanEval, MBPP) benchmarks, Shepherding reduces costs by 42-94% relative to LLM-only inference. Compared to state-of-the-art routing and cascading baselines, shepherding delivers up to 2.8x cost reduction while matching accuracy. To our knowledge, this is the first work to exploit token-level budget control for SLM-LLM collaboration.
format Preprint
id arxiv_https___arxiv_org_abs_2601_22132
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Pay for Hints, Not Answers: LLM Shepherding for Cost-Efficient Inference
Dong, Ziming
Sharma, Hardik
O'Toole, Evan
Champati, Jaya Prakash
Wu, Kui
Machine Learning
Large Language Models (LLMs) deliver state-of-the-art performance on complex reasoning tasks, but their inference costs limit deployment at scale. Small Language Models (SLMs) offer dramatic cost savings yet lag substantially in accuracy. Existing approaches - routing and cascading - treat the LLM as an all-or-nothing resource: either the query bypasses the LLM entirely, or the LLM generates a complete response at full cost. We introduce LLM Shepherding, a framework that requests only a short prefix (a hint) from the LLM and provides it to SLM. This simple mechanism is surprisingly effective for math and coding tasks: even hints comprising 10-30% of the full LLM response improve SLM accuracy significantly. Shepherding generalizes both routing and cascading, and it achieves lower cost under oracle decision-making. We develop a two-stage predictor that jointly determines whether a hint is needed and how many tokens to request. On the widely-used mathematical reasoning (GSM8K, CNK12) and code generation (HumanEval, MBPP) benchmarks, Shepherding reduces costs by 42-94% relative to LLM-only inference. Compared to state-of-the-art routing and cascading baselines, shepherding delivers up to 2.8x cost reduction while matching accuracy. To our knowledge, this is the first work to exploit token-level budget control for SLM-LLM collaboration.
title Pay for Hints, Not Answers: LLM Shepherding for Cost-Efficient Inference
topic Machine Learning
url https://arxiv.org/abs/2601.22132