Research map / Large language models / Instruction and fine-tuning
LinguaMap: Which Layers of LLMs Speak Your Language and How to Tune Them?
J. Ben Tamo, Daniel Carlander-Reuterfelt, Jonathan Rubin, Oleg Poliannikov, Dezhi Hong, Mingxian Wang
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 only the final layers (selective SFT) | Qwen-3-32B · MMLU non-Business (code-switched) | language consistency: 8.35 -> 99.62 | Improved |
| fine-tune only the final layers (selective SFT) | Qwen-3-32B · MMLU non-Business (code-switched) | task performance: 60.51 -> 74.44 | Improved |
| fine-tune only the final layers (selective SFT) | Qwen-3-32B · XQuAD (English distractor) | language consistency: 41.99 -> 97.62 | Improved |
| fine-tune only the final layers (selective SFT) | Qwen-3-32B · XQuAD (English distractor) | task performance: 15.81 -> 18.05 | No clear effect |
| selective SFT instead of full-scope SFT | Qwen-3-32B · XQuAD (English distractor) | language consistency: 75.99 -> 97.62 | Improved |
| selective SFT instead of full-scope SFT | Qwen-3-32B · MGSM (monolingual) | task performance: 90.53 -> 86.80 | Hurt |
| select final layers rather than a random layer subset (ablation) | Qwen-3-32B · MGSM (monolingual) | task performance: 0.13 -> 86.80 | Improved |
| select final layers rather than a random layer subset (ablation) | Qwen-3-32B · MGSM (monolingual) | language consistency: 65.87 -> 99.20 | Improved |
| select final layers rather than a random layer subset (ablation) | Bloom-7.1B · MGSM (monolingual) | language consistency: 69.47 -> 100.00 | Improved |
| fine-tune only the final layers (selective SFT) | Bloom-7.1B · MMLU non-Business (code-switched) | language consistency: 29.49 -> 98.66 | Improved |
| switch from monolingual to code-switched prompts | Qwen-3-32B · MMLU | language consistency: 45.17 -> 8.35 | Hurt |
| switch from monolingual to code-switched prompts | Qwen-3-32B · MMLU | task accuracy: 51.77 -> 60.5 | Improved |
Closest papers by meaning
- How do Large Language Models Handle Multilingualism? (NeurIPS 2024 · similarity 0.90)
- Tuning LayerNorm in Attention: Towards Efficient Multi-Modal LLM Finetuning (ICLR 2024 · similarity 0.90)
- Enhancing Non-English Capabilities of English-Centric Large Language Models Through Deep Supervision Fine-Tuning (AAAI 2025 · similarity 0.89)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.88)
- INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledge (ICLR 2025 · similarity 0.88)
- Language Ranker: A Metric for Quantifying LLM Performance Across High and Low-Resource Languages (AAAI 2025 · similarity 0.88)
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.
