Saved in:
Bibliographic Details
Main Authors: Finke, Lennart, Sreedhara, Chandan, Dooms, Thomas, Allen, Mat, Zhang, Emerald, Rodriguez, Juan Diego, Nabeshima, Noa, Marshall, Thomas, Braun, Dan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.09184
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916770177089536
author Finke, Lennart
Sreedhara, Chandan
Dooms, Thomas
Allen, Mat
Zhang, Emerald
Rodriguez, Juan Diego
Nabeshima, Noa
Marshall, Thomas
Braun, Dan
author_facet Finke, Lennart
Sreedhara, Chandan
Dooms, Thomas
Allen, Mat
Zhang, Emerald
Rodriguez, Juan Diego
Nabeshima, Noa
Marshall, Thomas
Braun, Dan
contents We present SimpleStories, a large synthetic story dataset in simple language, consisting of 2 million samples each in English and Japanese. Through parameterizing prompts at multiple levels of abstraction, we achieve control over story characteristics at scale, inducing syntactic and semantic diversity. Ablations on a newly trained model suite show improved sample efficiency and model interpretability compared to the TinyStories dataset. We open-source all constituent parts of model creation, hoping to enable novel ways to study the end-to-end training process. As a byproduct, we move the frontier regarding the fewest-parameter language model that outputs grammatical natural language.
format Preprint
id arxiv_https___arxiv_org_abs_2504_09184
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Parameterized Synthetic Text Generation with SimpleStories
Finke, Lennart
Sreedhara, Chandan
Dooms, Thomas
Allen, Mat
Zhang, Emerald
Rodriguez, Juan Diego
Nabeshima, Noa
Marshall, Thomas
Braun, Dan
Computation and Language
Artificial Intelligence
We present SimpleStories, a large synthetic story dataset in simple language, consisting of 2 million samples each in English and Japanese. Through parameterizing prompts at multiple levels of abstraction, we achieve control over story characteristics at scale, inducing syntactic and semantic diversity. Ablations on a newly trained model suite show improved sample efficiency and model interpretability compared to the TinyStories dataset. We open-source all constituent parts of model creation, hoping to enable novel ways to study the end-to-end training process. As a byproduct, we move the frontier regarding the fewest-parameter language model that outputs grammatical natural language.
title Parameterized Synthetic Text Generation with SimpleStories
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2504.09184