AI Paper Atlas

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

Reasoning Quality Emerges Early: Data Curation for Reasoning Models

Hongyi Henry Jin, Wenhan Yang, Meysam Ghaffari, Carlos Morato, Baharan Mirzasoleiman

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 0 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
add difficulty filtering using first-100-token loss at a randomly perturbed checkpoint (ablation)TEMP · M23k, Qwen2.5-7B-Instruct, 1kaverage accuracy (%): 54.09 -> 55.02Improved
add source budget weighting toward harder sources (ablation)TEMP · M23k, Qwen2.5-7B-Instruct, 1kaverage accuracy (%): 55.02 -> 55.28Improved
add diversity sampling via clustering of perturbed loss vectors (ablation)TEMP · M23k, Qwen2.5-7B-Instruct, 1kaverage accuracy (%): 55.28 -> 56.52Improved
add brittle example selection within clusters (ablation)TEMP · M23k, Qwen2.5-7B-Instruct, 1kaverage accuracy (%): 56.52 -> 57.69Improved
use an undertrained fine-tuned checkpoint to compute the fine-tuning direction (ablation)TEMP · M23k and OpenThoughts-MathaccuracyHurt
use an inexact but reasonable fine-tuned checkpoint for direction (ablation)TEMP · M23k and OpenThoughts-MathaccuracyNo clear effect
vary number of data picked per cluster k (ablation)TEMP · M23kaverage accuracyNo clear effect
cluster on first 1k reasoning tokens instead of the full response (ablation)TEMP · M23kaverage accuracyNo clear effect
use 1 or 2 perturbed checkpoints for difficulty filtering (ablation)TEMP · M23kaverage accuracyNo clear effect
use TEMP instead of S2L, learnability, perplexity, embedding diversity, M1k or randomSFT data selection · M23k, Llama3.1-8B-Instruct, 1kaverage accuracy (%): up to +1.7%Improved
use S2L loss-based selection on Qwen2.5SFT data selection · M23k, Qwen2.5-7B-Instruct, 1kaverage accuracyNo clear effect
compute losses only on initial tokens rather than full tracesTEMP · OpenThoughts-Math selectiontokens processed (million): 352.9 -> 30.3Improved

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