Adaptive Inference-Time Compute: LLMs Can Predict if They Can Do Better, Even Mid-Generation
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 reserved pith:EMDECOPWrecord.jsonopen to challenge →
read the original abstract
Inference-time computation is a powerful paradigm to enhance the performance of large language models (LLMs), with Best-of-N sampling being a widely used technique. However, this method is computationally expensive, requiring both (1) an external reward model and (2) the generation of multiple samples. In this work, we introduce a new generative self-evaluation scheme designed to adaptively reduce the number of generated samples while maintaining or even improving performance. We use a generative reward model formulation, allowing the LLM to predict mid-generation the probability that restarting the generation will yield a better response. These predictions are obtained without an external reward model and can be used to decide whether or not to generate more samples, prune unpromising samples early on, or to pick the best sample. This capability is very inexpensive as it involves generating a single predefined token. Trained using a dataset constructed with real unfiltered LMSYS user prompts, Llama 3.1 8B's win rate against GPT-4 on AlpacaEval increases from 21% to 34% with 16 samples and math performance on GSM8K improves from 84% to 91%. By sampling only when the LLM determines that it is beneficial to do so and adaptively adjusting temperature annealing, we demonstrate that 74% of the improvement from using 16 samples can be achieved with only 1.2 samples on average. We further demonstrate that 50-75% of samples can be pruned early in generation with minimal degradation in performance. Overall, our methods enable more efficient and scalable compute utilization during inference for LLMs.
This paper has not been read by Pith yet.
Forward citations
Cited by 9 Pith papers
-
Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade
A cascade of recall-calibrated gates on LLM agent hidden states aborts doomed episodes early, saving up to 47% compute at a 90% global success-recall target.
-
Hard or Just Unreached? Diagnosing the Sampling Blind Spot in Math-Reasoning Difficulty Estimation
10.3-22.9% of pass@k=0 math examples across GSM8K and MATH are recovered by a deterministic six-chain regime using activation grafting, showing a sampling blind spot in difficulty estimation.
-
ATLAS: Agentic Test-time Learning-to-Allocate Scaling
ATLAS introduces an LLM-orchestrated agentic framework for dynamic test-time scaling via extensible 'explore' actions, achieving higher accuracy with fewer API calls than fixed-workflow baselines on four benchmarks.
-
ModeX: Evaluator-Free Best-of-N Selection for Open-Ended Generation
ModeX selects the modal semantic output from multiple LLM generations via a similarity graph and recursive spectral clustering without needing reward models or evaluators.
-
Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
o1-like models overthink easy tasks; self-training reduces compute use without accuracy loss on GSM8K, MATH500, GPQA, and AIME.
-
AVIS: Adaptive Test-Time Scaling for Vision-Language Models
AVIS is an adaptive policy that jointly scales visual context via key-based token pruning and reasoning via difficulty-predicted self-consistency to improve the accuracy-compute curve on image and video tasks.
-
Adaptive Test-Time Compute Allocation with Evolving In-Context Demonstrations
An adaptive test-time framework uses a warm-up phase on the test set to build evolving in-context examples, then concentrates compute on unresolved queries to outperform static baselines on math, coding, and reasoning...
-
Explicit Reasoning Makes Better Judges: A Systematic Study on Accuracy, Efficiency, and Robustness
Thinking LLMs achieve ~10 percentage points higher accuracy than non-thinking ones on RewardBench with under 2x compute overhead, outperforming augmentation strategies that cost over 8x more while also showing better ...
-
A Survey of Scaling in Large Language Model Reasoning
A survey categorizing scaling in LLM reasoning across input size, steps, rounds, training, and future directions, noting that scaling can negatively affect performance.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.