What Do Learning Dynamics Reveal About Generalization in LLM Reasoning?
- URL: http://arxiv.org/abs/2411.07681v2
- Date: Mon, 18 Nov 2024 18:49:59 GMT
- Title: What Do Learning Dynamics Reveal About Generalization in LLM Reasoning?
- Authors: Katie Kang, Amrith Setlur, Dibya Ghosh, Jacob Steinhardt, Claire Tomlin, Sergey Levine, Aviral Kumar,
- Abstract summary: We find that a model's generalization behavior can be effectively characterized by a training metric we call pre-memorization train accuracy.
By connecting a model's learning behavior to its generalization, pre-memorization train accuracy can guide targeted improvements to training strategies.
- Score: 83.83230167222852
- License:
- Abstract: Despite the remarkable capabilities of modern large language models (LLMs), the mechanisms behind their problem-solving abilities remain elusive. In this work, we aim to better understand how the learning dynamics of LLM finetuning shapes downstream generalization. Our analysis focuses on reasoning tasks, whose problem structure allows us to distinguish between memorization (the exact replication of reasoning steps from the training data) and performance (the correctness of the final solution). We find that a model's generalization behavior can be effectively characterized by a training metric we call pre-memorization train accuracy: the accuracy of model samples on training queries before they begin to copy the exact reasoning steps from the training set. On the dataset level, this metric is able to reliably predict test accuracy, achieving $R^2$ of around or exceeding 0.9 across various models (Llama3 8, Gemma2 9B), datasets (GSM8k, MATH), and training configurations. On a per-example level, this metric is also indicative of whether individual model predictions are robust to perturbations in the training query. By connecting a model's learning behavior to its generalization, pre-memorization train accuracy can guide targeted improvements to training strategies. We focus on data curation as an example, and show that prioritizing examples with low pre-memorization accuracy leads to 1.5-2x improvements in data efficiency compared to i.i.d. data scaling, and outperforms other standard data curation techniques.
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) - Dynamic Loss-Based Sample Reweighting for Improved Large Language Model Pretraining [55.262510814326035]
Existing reweighting strategies primarily focus on group-level data importance.
We introduce novel algorithms for dynamic, instance-level data reweighting.
Our framework allows us to devise reweighting strategies deprioritizing redundant or uninformative data.
arXiv Detail & Related papers (2025-02-10T17:57:15Z) - Experience of Training a 1.7B-Parameter LLaMa Model From Scratch [10.39475177812483]
We share insights gained from training DMaS-LLaMa-Lite on approximately 20 billion tokens of data.
We chronicle the full training trajectory, documenting how evolving validation loss levels and downstream benchmarks reflect transitions from incoherent text to fluent, contextually grounded output.
By detailing these experiences and offering training logs, checkpoints, and sample outputs, we aim to guide future researchers and practitioners in refining their pretraining strategies.
arXiv Detail & Related papers (2024-12-17T21:15:52Z) - TaskMet: Task-Driven Metric Learning for Model Learning [29.0053868393653]
Deep learning models are often deployed in downstream tasks that the training procedure may not be aware of.
We propose take the task loss signal one level deeper than the parameters of the model and use it to learn the parameters of the loss function the model is trained on.
This approach does not alter the optimal prediction model itself, but rather changes the model learning to emphasize the information important for the downstream task.
arXiv Detail & Related papers (2023-12-08T18:59:03Z) - Learning Sample Difficulty from Pre-trained Models for Reliable
Prediction [55.77136037458667]
We propose to utilize large-scale pre-trained models to guide downstream model training with sample difficulty-aware entropy regularization.
We simultaneously improve accuracy and uncertainty calibration across challenging benchmarks.
arXiv Detail & Related papers (2023-04-20T07:29:23Z) - Theoretical Characterization of the Generalization Performance of
Overfitted Meta-Learning [70.52689048213398]
This paper studies the performance of overfitted meta-learning under a linear regression model with Gaussian features.
We find new and interesting properties that do not exist in single-task linear regression.
Our analysis suggests that benign overfitting is more significant and easier to observe when the noise and the diversity/fluctuation of the ground truth of each training task are large.
arXiv Detail & Related papers (2023-04-09T20:36:13Z) - Architecture, Dataset and Model-Scale Agnostic Data-free Meta-Learning [117.48444197402858]
We propose ePisode cUrriculum inveRsion (ECI) during data-free meta training and invErsion calibRation following inner loop (ICFIL) during meta testing.
ECI adaptively increases the difficulty level of pseudo episodes according to the real-time feedback of the meta model.
We formulate the optimization process of meta training with ECI as an adversarial form in an end-to-end manner.
arXiv Detail & Related papers (2023-03-20T15:10:41Z) - Harnessing the Power of Explanations for Incremental Training: A
LIME-Based Approach [6.244905619201076]
In this work, model explanations are fed back to the feed-forward training to help the model generalize better.
The framework incorporates the custom weighted loss with Elastic Weight Consolidation (EWC) to maintain performance in sequential testing sets.
The proposed custom training procedure results in a consistent enhancement of accuracy ranging from 0.5% to 1.5% throughout all phases of the incremental learning setup.
arXiv Detail & Related papers (2022-11-02T18:16:17Z) - Unrolling SGD: Understanding Factors Influencing Machine Unlearning [17.6607904333012]
Machine unlearning is the process through which a deployed machine learning model forgets about one of its training data points.
We first taxonomize approaches and metrics of approximate unlearning.
We identify verification error, i.e., the L2 difference between the weights of an approximately unlearned and a naively retrained model.
arXiv Detail & Related papers (2021-09-27T23:46:59Z)
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.