Smaller, Weaker, Yet Better: Training LLM Reasoners via Compute-Optimal Sampling
read the original abstract
Training on high-quality synthetic data from strong language models (LMs) is a common strategy to improve the reasoning performance of LMs. In this work, we revisit whether this strategy is compute-optimal under a fixed inference budget (e.g., FLOPs). To do so, we investigate the trade-offs between generating synthetic data using a stronger but more expensive (SE) model versus a weaker but cheaper (WC) model. We evaluate the generated data across three key metrics: coverage, diversity, and false positive rate, and show that the data from WC models may have higher coverage and diversity, but also exhibit higher false positive rates. We then finetune LMs on data from SE and WC models in different settings: knowledge distillation, self-improvement, and a novel weak-to-strong improvement setup where a weaker LM teaches reasoning to a stronger LM. Our findings reveal that models finetuned on WC-generated data consistently outperform those trained on SE-generated data across multiple benchmarks and multiple choices of WC and SE models. These results challenge the prevailing practice of relying on SE models for synthetic data generation, suggesting that WC may be the compute-optimal approach for training advanced LM reasoners.
This paper has not been read by Pith yet.
Forward citations
Cited by 6 Pith papers
-
Weak-to-Strong Generalization is Nearly Inevitable (in Linear Models)
Weak-to-strong generalization is nearly inevitable in linear logistic regression for most student-teacher pairs without any model capacity mismatch.
-
Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO
Smaller models provide temporally correlated policy-level diversity that serves as structured exploration for training larger models in GRPO, yielding accuracy gains such as +8.8% on AIME 24 with reduced compute via t...
-
Practical Scaling Laws: Converting Compute into Performance in a Data-Constrained World
A new scaling law L(N, D, T) = E + (L0 - E) h/(1+h) with h = a/N^α + b/T^β + c N^γ/D^δ that decomposes loss into undercapacity, undertraining, and overfitting terms and saturates between E and L0.
-
SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning
SingGuard presents a policy-adaptive multimodal LLM guardrail family with hybrid reasoning regimes and a new benchmark of 56,340 examples, claiming SOTA F1 across 35 datasets and improved policy adherence under runtim...
-
SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning
SingGuard introduces a policy-adaptive multimodal LLM guardrail with dynamic reasoning regimes and SingGuard-Bench, reporting SOTA F1 scores across 35 datasets and improved policy-following accuracy under runtime shifts.
-
Beyond Explainable AI (XAI): An Overdue Paradigm Shift and Post-XAI Research Directions
Current XAI methods for DNNs and LLMs rest on paradoxes and false assumptions that demand a paradigm shift to verification protocols, scientific foundations, context-aware design, and faithful model analysis rather th...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.