AI Paper Atlas

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

LangGPS: Language Separability Guided Data Pre-Selection for Joint Multilingual Instruction Tuning

Yangfan Ye, Xiaocheng Feng, Xiachong Feng, Lei Huang, Weitao Ma, Qichen Hong, Yunfei Lu, Duyu Tang, Dandan Tu, Bing Qin

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Instruction data selection · 2 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 LangGPS pre-selection to LESSLESS (LLaMA-3.1-8B) · six benchmarksrelative change vs full data: +4.90% -> +6.37%Improved
add LangGPS pre-selection to random selectionRand (LLaMA-3.1-8B) · XStoryClozeaccuracy: 55.4 -> 64.6Improved
add LangGPS pre-selection to LESSLESS (LLaMA-3.1-8B) · MKQAaccuracy: 17.8 -> 16.9Hurt
add LangGPS pre-selection to KMCKMC (Qwen2.5-7B) · XStoryClozeaccuracy: 70.6 -> 73.7Improved
add LangGPS pre-selection to DSIRDSIR (Qwen2.5-7B) · MMMLUaccuracy: 45.4 -> 48.7Improved
add LangGPS pre-selection to MTLDMTLD (Qwen2.5-7B) · XLSumROUGE-L: 22.2 -> 21.8Hurt
add LangGPS pre-selection to CohCoh (LLaMA-3.1-8B) · XQuADaccuracy: 63.2 vs 64.5Hurt
choose LESS over random selectionLESS (Qwen2.5-7B) · six benchmarksrelative change vs full data: -7.35% vs +4.18%Hurt
use balanced separability curriculum instead of random shuffle (ablation)full-data SFT (LLaMA-3.1-8B) · XStoryClozeaccuracy: 56.8 -> 68.8Improved
use balanced separability curriculum instead of random shuffle (ablation)full-data SFT (Qwen2.5-7B) · XNLIaccuracy: 45.3 -> 50.8Improved
use ascending separability curriculum instead of random shuffle (ablation)full-data SFT (Qwen2.5-7B) · XNLIaccuracy: 45.3 -> 36.4Hurt
vary pre-selection ratio rho from 10% to 100% (ablation)LangGPS+Rand/LESS (LLaMA-3.1-8B) · MMMLU, XLSum (5%)accuracy / ROUGE-LMixed

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