Rethinking Fine-Tuning when Scaling Test-Time Compute: Limiting Confidence Improves Mathematical Reasoning

• URL: http://arxiv.org/abs/2502.07154v1
• Date: Tue, 11 Feb 2025 00:33:31 GMT
• Title: Rethinking Fine-Tuning when Scaling Test-Time Compute: Limiting Confidence Improves Mathematical Reasoning
• Authors: Feng Chen, Allan Raventos, Nan Cheng, Surya Ganguli, Shaul Druckmann,
• Abstract summary: We show that training with cross-entropy loss can be misaligned with pass@N in that pass@N accuracy $it decreases$ with longer training. We suggest a principled, modified training loss that is better aligned to pass@N by limiting model confidence and rescuing pass@N test performance.
• Score: 32.45574194957491
• License:
• Abstract: Recent progress in large language models (LLMs) highlights the power of scaling test-time compute to achieve strong performance on complex tasks, such as mathematical reasoning and code generation. This raises a critical question: how should model training be modified to optimize performance under a subsequent test-time compute strategy and budget? To explore this, we focus on pass@N, a simple test-time strategy that searches for a correct answer in $N$ independent samples. We show, surprisingly, that training with cross-entropy (CE) loss can be ${\it misaligned}$ with pass@N in that pass@N accuracy ${\it decreases}$ with longer training. We explain the origins of this misalignment in terms of model overconfidence induced by CE, and experimentally verify our prediction of overconfidence as an impediment to scaling test-time compute via pass@N. Furthermore we suggest a principled, modified training loss that is better aligned to pass@N by limiting model confidence and rescuing pass@N test performance. Our algorithm demonstrates improved mathematical reasoning on MATH and MiniF2F benchmarks under several scenarios: (1) providing answers to math questions; and (2) proving theorems by searching over proof trees of varying shapes. Overall our work underscores the importance of co-designing two traditionally separate phases of LLM development: training-time protocols and test-time search and reasoning strategies.

Related papers

• S$^2$R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning [51.84977135926156]
  We introduce S$2$R, an efficient framework that enhances LLM reasoning by teaching models to self-verify and self-correct during inference. Our results demonstrate that Qwen2.5-math-7B achieves an accuracy improvement from 51.0% to 81.6%, outperforming models trained on an equivalent amount of long-CoT distilled data.
  arXiv  Detail & Related papers  (2025-02-18T13:40:22Z)
• Scaling Test-Time Compute Without Verification or RL is Suboptimal [70.28430200655919]
  We show that finetuning LLMs with verifier-based (VB) methods based on RL or search is far superior to verifier-free (VF) approaches based on distilling or cloning search traces, given a fixed amount of compute/data budget. We corroborate our theory empirically on both didactic and math reasoning problems with 3/8B-sized pre-trained LLMs, where we find verification is crucial for scaling test-time compute.
  arXiv  Detail & Related papers  (2025-02-17T18:43:24Z)
• Learning to Stop Overthinking at Test Time [1.0356759327536202]
  Test time scaling is one of the most active research areas that shows promise after training time scaling has reached its limits. We introduce a test time training method for determining the optimal amount of computation needed for each sample during test time. We also propose Conv-LiGRU, a novel recurrent architecture for efficient and robust visual reasoning.
  arXiv  Detail & Related papers  (2025-02-16T02:17:05Z)
• Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs [76.43407125275202]
  o1-like models can emulate human-like long-time thinking during inference. This paper presents the first comprehensive study on the prevalent issue of overthinking in these models. We propose strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy.
  arXiv  Detail & Related papers  (2024-12-30T18:55:12Z)
• Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters [27.656263126925815]
  We study the scaling of inference-time computation in LLMs. We find that in both cases, the effectiveness of different approaches to scaling test-time compute critically varies depending on the difficulty of the prompt.
  arXiv  Detail & Related papers  (2024-08-06T17:35:05Z)
• Test-Time Adaptation with Perturbation Consistency Learning [32.58879780726279]
  We propose a simple test-time adaptation method to promote the model to make stable predictions for samples with distribution shifts. Our method can achieve higher or comparable performance with less inference time over strong PLM backbones.
  arXiv  Detail & Related papers  (2023-04-25T12:29:22Z)
• Improving Representational Continuity via Continued Pretraining [76.29171039601948]
  Transfer learning community (LP-FT) outperforms naive training and other continual learning methods. LP-FT also reduces forgetting in a real world satellite remote sensing dataset (FMoW) variant of LP-FT gets state-of-the-art accuracies on an NLP continual learning benchmark.
  arXiv  Detail & Related papers  (2023-02-26T10:39:38Z)
• Confident Adaptive Language Modeling [95.45272377648773]
  CALM is a framework for dynamically allocating different amounts of compute per input and generation timestep. We demonstrate the efficacy of our framework in reducing compute -- potential speedup of up to $times 3$ -- while provably maintaining high performance.
  arXiv  Detail & Related papers  (2022-07-14T17:00:19Z)
• Test-time Batch Normalization [61.292862024903584]
  Deep neural networks often suffer the data distribution shift between training and testing. We revisit the batch normalization (BN) in the training process and reveal two key insights benefiting test-time optimization. We propose a novel test-time BN layer design, GpreBN, which is optimized during testing by minimizing Entropy loss.
  arXiv  Detail & Related papers  (2022-05-20T14:33:39Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.