How Truncating Weights Improves Reasoning in Language Models
In addition to the ability to generate fluent text in various languages, large language models have been successful at tasks that involve basic forms of logical "reasoning" over their context. Recent work found that selectively removing certain components from weight matrices in pre-trained models can improve such reasoning capabilities. We investigate this phenomenon further by carefully studying how certain global associations tend to be stored in specific weight components or Transformer blocks, in particular feed-forward layers. Such associations may hurt predictions in reasoning tasks, and removing the corresponding components may then improve performance. We analyze how this arises during training, both empirically and theoretically, on a two-layer Transformer trained on a basic reasoning task with noise, a toy associative memory model, and on the Pythia family of pre-trained models tested on simple reasoning tasks.
Code (0)
등록된 구현이 없습니다.
Tasks
Logical ReasoningMethods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
Power-boosting in Specification Tests using Kernel Directional Component
We propose power-boosting strategies for kernel-based specification tests in conditional moment models, with a focus on the Kernel Conditional Moment (KCM) test. By decomposing the KCM statistic into spectral components,…
WS-GRPO: Weakly-Supervised Group-Relative Policy Optimization for Rollout-Efficient Reasoning
Group Relative Policy Optimization (GRPO) is effective for training language models on complex reasoning. However, since the objective is defined relative to a group of sampled trajectories, extended deliberation can cre…
Bias-Restrained Prefix Representation Finetuning for Mathematical Reasoning
Parameter-Efficient finetuning (PEFT) enhances model performance on downstream tasks by updating a minimal subset of parameters. Representation finetuning (ReFT) methods further improve efficiency by freezing model weigh…
Mathematical ReasoningKnowing Before Saying: LLM Representations Encode Information About Chain-of-Thought Success Before Completion
We investigate whether the success of a zero-shot Chain-of-Thought (CoT) process can be predicted before completion. We discover that a probing classifier, based on LLM representations, performs well \emph{even before a …
Adaptively Weighted Data Augmentation Consistency Regularization for Robust Optimization under Concept Shift
Concept shift is a prevailing problem in natural tasks like medical image segmentation where samples usually come from different subpopulations with variant correlations between features and labels. One common type of co…
Data AugmentationImage SegmentationMedical Image SegmentationSegmentation+2