Research map / Large language models / Instruction and fine-tuning
MIDB: Multilingual Instruction Data Booster for Enhancing Cultural Equality in Multilingual Instruction Synthesis
Yilun Liu, Chunguang Zhao, Xinhua Yang, Hongyong Zeng, Shimin Tao, Weibin Meng, Minggui He, Yan Yu, Hongxia Ma, Li Zhang, Daimeng Wei, Boxing Chen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune LLaMA3-8B on MIDB-boosted multilingual Alpaca instead of machine-translated Alpaca | Alpaca-Original · BLEnD cultural knowledge, Korean | accuracy: 18.50 -> 24.19 | Improved |
| fine-tune on MIDB-boosted data | Alpaca-Original · BLEnD cultural knowledge, Arabic | accuracy: 15.03 -> 16.85 | Improved |
| fine-tune on MIDB-boosted data | Alpaca-Original · AlpacaEval-16L human judged, Portuguese | winning score: 1.62 vs original (score >1 is win) | Improved |
| fine-tune on MIDB-boosted data | Alpaca-Original · MT-Bench-16L round 1 human, Korean | winning score: 1.38 (barely above 1) | No clear effect |
| apply MIDB boosting to machine-translated Alpaca data | original data · data quality, Portuguese (LLM judge) | win ratio: 46% win vs 9% lose | Improved |
| apply MIDB boosting to out-of-distribution Dolly-15k | original data · Dolly-15k data quality | LLM-judge win ratio | Improved |
| use Qwen2.5-7B/Qwen3-8B instead of LLaMA3.1-8B as MIDB backbone (ablation) | MIDB · data quality winning score | winning score | Hurt |
| train MIDB on all revision samples instead of top 30% most revised (alpha) (ablation) | MIDB · data quality winning score | winning score | Hurt |
Closest papers by meaning
- XIFBench: Evaluating Large Language Models on Multilingual Instruction Following (NeurIPS 2025 · similarity 0.89)
- Teaching Large Language Models to Translate with Comparison (AAAI 2024 · similarity 0.88)
- SHED: Shapley-Based Automated Dataset Refinement for Instruction Fine-Tuning (NeurIPS 2024 · similarity 0.87)
- What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuning (ICLR 2024 · similarity 0.87)
- T-SHIRT: Token-Selective Hierarchical Data Selection for Instruction Tuning (NeurIPS 2025 · similarity 0.86)
- Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuning (ICML 2024 · similarity 0.86)
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.
