AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Revisiting the Scaling Properties of Downstream Metrics in Large Language Model Training

Jakub Krajewski, Amitis Shidani, Dan Busbridge, Sam Wiseman, Jason Ramapuram

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 6 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
predict accuracy directly from FLOPs with a power law instead of a two-stage linear NLL proxyaccuracy prediction · mean over benchmarksMRE (%): 6.667 -> 1.963Improved
predict directly with a power law instead of two-stage logisticaccuracy prediction · mean over benchmarksMRE (%): 6.351 -> 1.963Improved
use power law instead of BNSLaccuracy prediction · mean over benchmarksMRE (%): 2.713 -> 1.963Improved
use power law instead of BNSLaccuracy prediction · average over 12 benchmarksMRE (%): 4.96 -> 4.72Improved
use BNSL instead of power lawaccuracy prediction · WebQSMRE (%): 7.28 -> 1.46Improved
use BNSL instead of power lawaccuracy prediction · LBPPMRE (%): 13.43 -> 6.70Improved
use BNSL instead of power lawaccuracy prediction · GSM8KMRE (%): 9.67 -> 13.90Hurt
choose a different proxy metric in two-stage (log-prob, loss, Brier) (ablation)two-stage scaling law · benchmarksR2No clear effect
two-stage models fit training data better yet extrapolate worsetwo-stage logistic · training fitR2: 0.974 vs 0.986Mixed
fit only on runs with accuracy at least 5 points above random (ablation)power law fit · benchmarksfit qualityImproved
change pretraining data from the main mixture to C4power law scaling · code and math tasksaccuracyHurt
change pretraining data to C4power law scaling · general QA tasksfit qualityNo clear effect

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