AI Paper Atlas

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

BLISS: A Lightweight Bilevel Influence Scoring Method for Data Selection in Language Model Pretraining

Jie Hao, Rui Yu, Wei Zhang, Huixia Wang, Jie Xu, Mingrui Liu

ICML 2026 · Large language models › Language, safety and interpretability · approach: Pruning and circuits in LLMs · 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 softmax reparametrization of score outputs (ablation)BLISS · Zero-shot average, 410M, 10B tokens, 20k stepsaverage accuracy: 43.6 -> 44.7Improved
use bilevel instead of single-level update (ablation)BLISS · Zero-shot average, 410M, 10B tokens, 20k stepsaverage accuracy: 44.2 -> 44.7Improved
use 160M proxy instead of 31M (ablation)BLISS · Zero-shot average, 410M, 10B tokens, 20k stepsaverage accuracy: 44.7 -> 44.1No clear effect
initialize score model from previous round rather than round-1 reset (ablation)BLISS · Zero-shot average, 410M, 15B tokensaverage accuracy: 44.5 -> 44.9Improved
replace MATES with BLISSPythia-410M · Zero-shot average, 9 tasks, C4 25B tokensaverage accuracy: 45.7 -> 45.9Improved
replace MATES with BLISSPythia-1B · Zero-shot average, 9 tasks, C4 25B tokensaverage accuracy: 47.5 -> 47.9Improved
replace MATES with BLISS at round 3Pythia-2.8B · Round 3, 15B tokensaverage accuracy: 47.6 -> 49.0Improved
replace MATES with BLISS at round 1Pythia-2.8B · Round 1 (random), 15B tokensaverage accuracy: 45.9 -> 45.2No clear effect
replace MATES with BLISS at round 3LLaMA-0.5B · LLaMA-0.5B Round 3average accuracy: 45.01 -> 45.65Improved
replace MATES with BLISS for selection timePythia-410M · Data selection time, target 410Mhours: 18.0 -> 5.03Improved
replace MATES with BLISS for memoryPythia-1B · Data selection memory, target 1Bmemory peak (GB): 63.52 -> 74.51Hurt
increase bilevel steps from 3k to 5k (ablation)BLISS · Pythia proxy training losstraining lossNo clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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