AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Scaling Data-Constrained Language Models

Niklas Muennighoff, Alexander M Rush, Boaz Barak, Teven Le Scao, Nouamane Tazi, Aleksandra Piktus, Sampo Pyysalo, Thomas Wolf, Colin Raffel

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws · 32 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train up to 4 epochs on repeated data instead of unique data8.7B GPT-2 style LM · C4 validation lossvalidation loss: 0.5% higher validation lossNo clear effect
allocate compute by data-constrained law (fewer params, more epochs) vs Chinchilla allocation9.3e21 FLOPs budget, 25B unique tokens · C4 test lossloss: 27% fewer parameters, better lossImproved
fill up to 50% of data budget with Python code4.2B LM, 84B tokens · 19 NL tasks averagedownstream scoreNo clear effect
fill more than 50% of data with code4.2B LM, 84B tokens · 19 NL tasks averagedownstream scoreHurt
perplexity filtering (top 25% lowest perplexity)4.2B LM, 84B tokens · 19 NL tasks averagedownstream scoreImproved
deduplication filtering4.2B LM, 84B tokens · 19 NL tasks averagedownstream scoreNo clear effect
repeat data beyond about 4 epochs4.2B LM · downstream averagedownstream scoreHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy