CurLL: A Developmental Framework to Evaluate Continual Learning in Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kalyan, Pavan, Mishra, Shubhra, Lokam, Satya, Goyal, Navin
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914093660635136
author Kalyan, Pavan
Mishra, Shubhra
Lokam, Satya
Goyal, Navin
author_facet Kalyan, Pavan
Mishra, Shubhra
Lokam, Satya
Goyal, Navin
contents We introduce a comprehensive continual learning dataset and benchmark (CurlL) grounded in human developmental trajectories from ages 5-10, enabling systematic and fine-grained assessment of models' ability to progressively acquire new skills. CurlL spans five developmental stages (0-4) covering ages 5-10, supported by a skill graph that breaks down broad skills into smaller abilities, concrete goals, and measurable indicators, while also capturing which abilities build on others. We generate a 23.4B-token synthetic dataset with controlled skill progression, vocabulary complexity, and format diversity, comprising paragraphs, comprehension-based QA (CQA), skill-testing QA (CSQA), and instruction-response (IR) pairs. Stage-wise token counts range from 2.12B to 6.78B tokens, supporting precise analysis of forgetting, forward transfer, and backward transfer. Using a 135M-parameter transformer trained under independent, joint, and sequential (continual) setups, we show trade-offs in skill retention and transfer efficiency. By mirroring human learning patterns and providing fine-grained control over skill dependencies, this work advances continual learning evaluations for language models.
format Preprint
id arxiv_https___arxiv_org_abs_2510_13008
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CurLL: A Developmental Framework to Evaluate Continual Learning in Language Models
Kalyan, Pavan
Mishra, Shubhra
Lokam, Satya
Goyal, Navin
Computation and Language
Artificial Intelligence
We introduce a comprehensive continual learning dataset and benchmark (CurlL) grounded in human developmental trajectories from ages 5-10, enabling systematic and fine-grained assessment of models' ability to progressively acquire new skills. CurlL spans five developmental stages (0-4) covering ages 5-10, supported by a skill graph that breaks down broad skills into smaller abilities, concrete goals, and measurable indicators, while also capturing which abilities build on others. We generate a 23.4B-token synthetic dataset with controlled skill progression, vocabulary complexity, and format diversity, comprising paragraphs, comprehension-based QA (CQA), skill-testing QA (CSQA), and instruction-response (IR) pairs. Stage-wise token counts range from 2.12B to 6.78B tokens, supporting precise analysis of forgetting, forward transfer, and backward transfer. Using a 135M-parameter transformer trained under independent, joint, and sequential (continual) setups, we show trade-offs in skill retention and transfer efficiency. By mirroring human learning patterns and providing fine-grained control over skill dependencies, this work advances continual learning evaluations for language models.
title CurLL: A Developmental Framework to Evaluate Continual Learning in Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2510.13008