Input-Time Scaling
2025/08/19 by Huang, Rapheal, Guo, Weilong
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG)
paper · doi:10.48550/arxiv.2508.13654
Abstract
Current Large Language Models (LLMs) are usually post-trained on large-scale carefully curated datasets (data & training scaling) and doing reasoning in test time (inference time scaling). In this work, we present a new scaling paradigm, Input-Time Scaling, to complement previous scaling methods by putting resources on queries (input time). During training and testing, we utilize meta-knowledge from LLMs to refine inputs with different strategies. We also discover a new phenomenon, train-test co-design. It requires us to apply query strategies during training and testing as a whole. Only applying strategies on training or testing would seriously degrade the performance gained. We are also surprised to find that seemingly low data quality datasets can perform better. We can get the best performance even by adding irrelevant information to the queries, with randomly selected 1k examples from a minimally filtered dataset. These findings contradict the widely held inductive bias, "garbage in, garbage out". Curating datasets with seemingly high-quality data can even potentially limit the performance ceiling. In addition, models trained on more data with similar quality (15k VS 1k) perform worse, the intuition of simply scaling the size should also be carefully inspected. The good news is that our findings are compatible with the Less is More phenomenon. 1K examples are enough to invoke high-level reasoning ability. With experiments on Qwen2.5-32B-Instruct, we are able to reach SOTA performance among 32B models on AIME24(76.7%) and AIME25(76.7%) pass@1. We can further achieve AIME24(76.7%) and AIME25(80%) with a majority vote of three models. Starting from DeepSeek-R1-Distill-Qwen-32B, the result would be 90.0% on AIME24 and 80.0% on AIME25. To facilitate reproducibility and further research, we are working on open-source our datasets, data pipelines, evaluation results, and checkpoints.
Citations
- CTTS: Collective Test-Time Scaling
- Diversity-Enhanced Reasoning for Subjective Questions
- MiroMind-M1: An Open-Source Advancement in Mathematical Reasoning via Context-Aware Multi-Stage Policy Optimization
- Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiency
- Infinity Instruct: Scaling Instruction Selection and Synthesis to Enhance Language Models
- OpenThoughts: Data Recipes for Reasoning Models
- Reflect, Retry, Reward: Self-Improving LLMs via Reinforcement Learning
- Skywork Open Reasoner 1 Technical Report
- Reasoning Model is Stubborn: Diagnosing Instruction Overriding in Reasoning Models
- Grokking in the Wild: Data Augmentation for Real-World Multi-Hop Reasoning with Transformers
- LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities
- Climbing the Ladder of Reasoning: What LLMs Can-and Still Can't-Solve after SFT?
- Inference-Time Scaling for Complex Tasks: Where We Stand and What Lies Ahead
- Revisiting the Test-Time Scaling of o1-like Models: Do they Truly Possess Test-Time Scaling Capabilities?
- On the Effect of Sampling Diversity in Scaling LLM Inference
- LIMO: Less is More for Reasoning
- Rethinking Mixture-of-Agents: Is Mixing Different Large Language Models Beneficial?
- s1: Simple test-time scaling
- Surveying the Effects of Quality, Diversity, and Complexity in Synthetic Data From Large Language Models
- Only-IF:Revealing the Decisive Effect of Instruction Diversity on Generalization
- Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning
- Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
- The Llama 3 Herd of Models
- Qwen2 Technical Report
- Scaling Synthetic Data Creation with 1,000,000,000 Personas
- Metacognitive Capabilities of LLMs: An Exploration in Mathematical Problem Solving
- LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models
- Arcee's MergeKit: A Toolkit for Merging Large Language Models
- GPQA: A Graduate-Level Google-Proof Q&A Benchmark
- Self-Consistency Improves Chain of Thought Reasoning in Language Models
- BNAI, NO-TOKEN, and MIND-UNITY: Pillars of a Systemic Revolution in Artificial Intelligence
- Measuring Mathematical Problem Solving With the MATH Dataset
Related