AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Train-before-Test Harmonizes Language Model Rankings

Guanhua Zhang, Ricardo Dominguez-Olmedo, Moritz Hardt

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 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
fine-tune each model on the benchmark train set before testing61 LLMs, 24 benchmarks · Cross-benchmark ranking agreementmean Kendall tau: 0.52 -> 0.76Improved
fine-tune before testing, ties by statistical significance61 LLMs · Cross-benchmark ranking agreementmean Kendall tau-b: 0.58 -> 0.77Improved
fine-tune before testingNQ-Open outlier benchmark · Agreement with other benchmarksmean Kendall tau: 0.23 -> 0.74Improved
fine-tune before testinglanguage understanding category · Intra-category agreementmean Kendall tau: 0.52 -> 0.75Improved
fine-tune before testingmath category · Intra-category agreementmean Kendall tau: 0.55 -> 0.84Improved
fine-tune before testingperplexity vs downstream rankings · Perplexity-downstream agreementmean Kendall tau: 0.48 -> 0.74Improved
fine-tune before testingaverage perplexity vs average downstream · Aggregate agreementKendall tau: 0.55 -> 0.84Improved
fine-tune before testingmodel score matrix, all models · PCAPC1 explained variance: 70% -> 86%Improved
restrict to Qwen family and fine-tune before testingQwen score matrix · PCAPC1 explained variance: 86% -> 93%Improved
fine-tune before testingperplexity-perplexity agreement across corpora · Perplexity ranking agreementmean Kendall tau: 0.76 -> 0.78No clear effect
use instruction-tuned instead of base models for perplexity predicting potentialpre-fine-tuning perplexity · Post-fine-tuning downstream rankingsmean Kendall tau: 0.78 -> 0.51Hurt
account for statistical significance (ties) in direct evaluationdirect evaluation rankings · Cross-benchmark ranking agreementmean Kendall tau: 0.52 -> 0.58Improved

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