Papers with Instruction tuning
LawInstruct: A Resource for Studying Language Model Adaptation to the Legal Domain (2025.findings-naacl)
Copied to clipboard
Joel Niklaus, Lucia Zheng, Arya D. McCarthy, Christopher Hahn, Brian M Rosen, Peter Henderson, Daniel E. Ho, Garrett Honke, Percy Liang, Christopher D Manning
| Challenge: | In general, instruction tuning is important for direct user interaction, but the legal domain is underrepresented in typical instruction datasets. |
| Approach: | They aggregate 58 annotated legal datasets and write instructions for each to create LawInstruct. |
| Outcome: | The proposed model improves on LegalBench across all model sizes, but no drop in MMLU. |
Call for Rigor in Reporting Quality of Instruction Tuning Data (2025.acl-short)
Copied to clipboard
| Challenge: | Instruction tuning is crucial for adapting large language models (LLMs) to user intentions. |
| Approach: | They propose to use hyperparameters for training models that are often selected arbitrarily without adequate justification to make arbitrary conclusions. |
| Outcome: | The results show that arbitrary hyperparameter decisions can make any arbitrary conclusion. |
Demystifying Instruction Mixing for Fine-tuning Large Language Models (2024.acl-srw)
Copied to clipboard
| Challenge: | Instruction tuning is effective for aligning large language models with human instructions, but the procedure to optimizing the mixing of instruction datasets is still unclear. |
| Approach: | They categorize instructions into three primary types: NLP downstream tasks, coding, and general chat. |
| Outcome: | The proposed method improves performance of large language models (LLMs) but it is difficult to combine different instruction datasets to optimize overall performance. |
InstructDial: Improving Zero and Few-shot Generalization in Dialogue through Instruction Tuning (2022.emnlp-main)
Copied to clipboard
| Challenge: | Instruction tuning is emerging in NLP, but has not been explored for dialogue-related tasks. |
| Approach: | They propose an instruction tuning framework for dialogue that leverages natural language instructions with language models to induce zero-shot generalization on unseen tasks. |
| Outcome: | The proposed framework enables good zero-shot performance on unseen datasets and tasks such as dialogue evaluation and intent detection. |
MergeIT: From Selection to Merging for Efficient Instruction Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning rely on LLMs to score instruction quality . existing methods rely only on Llms to rank instruction quality, but this approach is expensive and time-consuming . |
| Approach: | They propose a novel LLM-based Merging strategy for better Instruction Tuning that shifts the focus from selection to synthesis. |
| Outcome: | The proposed method reduces time and computational cost while preserving diversity and reducing redundancy. |
Optimizing Instruction Synthesis: Effective Exploration of Evolutionary Space with Tree Search (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Extensive research has highlighted the quality of instruction data is essential for the success of this alignment. |
| Approach: | They propose a framework for iteratively improving existing instruction data by using Monte Carlo tree search to find suitable prompts that align the language model to effectively learn multiple skills. |
| Outcome: | The proposed framework improves the evaluation scores of seed instruction data, raising the average evaluation scores from 2.19 to 3.81. |
A Simple-Yet-Efficient Instruction Augmentation Method for Zero-Shot Sentiment Classification (2025.coling-main)
Copied to clipboard
| Challenge: | Existing studies have used labeled sentiment instances to instruction tune LLMs, improving zero-shot sentiment classification performance. |
| Approach: | They propose a simple-yet-efficient method which does not rely on actual labeled sentiment instances. |
| Outcome: | The proposed method outperforms LLMs tuned with more complex instruction tuning methods by 5.1 points and increases scores by 30 points. |
Instruction Data Selection via Answer Divergence (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning use data-centric methods, but they do not explicitly reflect what a particular base model is missing. |
| Approach: | They propose a method for instruction tuning that uses geometric structure of multi-sample outputs to select instruction data. |
| Outcome: | The proposed approach outperforms strong selectors on six benchmarks spanning reasoning, knowledge, and coding. |
Med-MoE: Mixture of Domain-Specific Experts for Lightweight Medical Vision-Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in multimodal large language models have seen remarkable progress for medical decision-making, however, they are designated for specific classification or generative tasks and require model training or finetuning on large-scale datasets with sizeable parameters and tremendous computing. |
| Approach: | They propose a framework that tackles discriminative and generative multimodal medical tasks using multimodal alignment, instruction tuning and routing. |
| Outcome: | The proposed model can achieve superior performance to or on par with state-of-the-art baselines while only requiring 30%-50% of activated model parameters. |
Stronger Models are Not Always Stronger Teachers for Instruction Tuning (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to optimize instruction-following capabilities of large language models (LLMs) assume that larger or stronger models are stronger teachers and therefore adopt smaller models as response generators. |
| Approach: | They propose to use large-scale instruction datasets to tune large language models to align with specific tasks and user intents. |
| Outcome: | The proposed metric outperforms most baselines in identifying the effectiveness of response generators. |
CodecLM: Aligning Language Models with Tailored Synthetic Data (2024.findings-naacl)
Copied to clipboard
Zifeng Wang, Chun-Liang Li, Vincent Perot, Long Le, Jin Miao, Zizhao Zhang, Chen-Yu Lee, Tomas Pfister
| Challenge: | Recent work on generating diverse instructions and applying LLM to increase instruction complexity neglects downstream use cases. |
| Approach: | They propose a framework for generating high-quality synthetic data for LLM alignment with different downstream instruction distributions and LLMs. |
| Outcome: | Experiments on four open-domain instruction using the proposed framework validate the effectiveness of CodecLM over the current state-of-the-art. |
Dynosaur: A Dynamic Growth Paradigm for Instruction-Tuning Data Curation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning do not include associating instructions with existing datasets. |
| Approach: | They propose a dynamic growth paradigm for the automatic curation of instruction-tuning data . they use existing datasets to automatically construct instruction-uning datasets . |
| Outcome: | The proposed model reduces the API cost for generating instructions and provides high-quality data. |
TAGCOS: Task-agnostic Gradient Clustered Coreset Selection for Instruction Tuning Data (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning are limited due to the increasing volume of instruction datasets and the increased computational costs. |
| Approach: | They propose to extract a small and highly informative subset of training samples from a large dataset that achieves comparable performance to the full dataset. |
| Outcome: | The proposed algorithm outperforms other unsupervised methods and achieves comparable performance to the full dataset. |
Instruction Tuning with and without Context: Behavioral Shifts and Downstream Impact (2026.eacl-long)
Copied to clipboard
Hyunji Lee, Seunghyun Yoon, Yunjae Won, Hanseok Oh, Geewook Kim, Trung Bui, Franck Dernoncourt, Elias Stengel-Eskin, Mohit Bansal, Minjoon Seo
| Challenge: | Prior work on instruction tuning datasets combined these data types without examining their distinct effects. |
| Approach: | They investigate how training LLMs with or without context affects model behavior and performance . they find that using context-augmented data as the backbone for vision-language models reduces hallucination . |
| Outcome: | The proposed training with context-augmented data reduces hallucination and improves grounding in the visual domain. |
PLUG: Leveraging Pivot Language in Cross-Lingual Instruction Tuning (2024.acl-long)
Copied to clipboard
| Challenge: | Instruction tuning has advanced large language models (LLMs) but its application in lower-resource languages faces challenges due to the imbalanced foundational abilities of LLMs across different languages. |
| Approach: | They propose a pivot language guided generation approach that utilizes a high-resource language as the pivot to enhance instruction tuning in lower-resourced languages. |
| Outcome: | The proposed approach improves instruction-following abilities of LLMs by 29% on average compared to directly responding in the target language alone. |
Achieving Stronger Generation via Simple Contrastive Tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent years have witnessed remarkable progress in large language models (LLMs). |
| Approach: | They propose a framework for contrastive decoding to enhance instruction-tuned models. |
| Outcome: | The proposed framework improves model performance without additional data or computational resources. |
CRaSh: Clustering, Removing, and Sharing Enhance Fine-tuning without Full Large Language Model (2023.emnlp-main)
Copied to clipboard
| Challenge: | Instruction tuning is an effective way of aligning large language models with private instruction data. |
| Approach: | They propose a training-free strategy to derive improved emulators from LLMs by using Offsite-Tuning (OFT) they propose CRaSh, which transfers transformer blocks between centralized LLM and downstream emulators . |
| Outcome: | The proposed technique boosts performance of large language models with billions of parameters. |
Contrastive Instruction Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Current LLMs exhibit limited robustness to unseen instructions, generating inconsistent outputs when the same instruction is phrased with slightly varied forms or language styles. |
| Approach: | They propose a method which maximizes the similarity between the hidden representations of semantically equivalent instruction-instance pairs while minimizing the similarities between semantically different ones. |
| Outcome: | Experiments on the PromptBench benchmark show that Contrastive Instruction Tuning improves LLMs’ robustness to unseen instructions with variations across character, word, sentence, and semantic levels by +2.5% in accuracy. |
MAIN: Mutual Alignment Is Necessary for instruction tuning (2025.emnlp-main)
Copied to clipboard
Fanyi Yang, Jianfeng Liu, Xin Zhang, Haoyu Liu, Xixin Cao, Yuefeng Zhan, Hao Sun, Weiwei Deng, Feng Sun, Qi Zhang
| Challenge: | Instruction tuning has enabled large language models to achieve remarkable performance, yet its success heavily depends on the availability of high-quality instruction-response pairs. |
| Approach: | They propose a mutual alignment framework which enforces coherence between instructions and responses through mutual constraints. |
| Outcome: | The proposed framework generalizes well across model architectures and sizes, achieving state-of-the-art performance on LLaMA, Mistral, and Qwen models across diverse benchmarks. |
HiDe-LLaVA: Hierarchical Decoupling for Continual Instruction Tuning of Multimodal Large Language Model (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods to improve instructionfollowing performance of MLLMs often trade off memory efficiency for performance gains, compromising overall efficiency. |
| Approach: | They propose a task-specific expansion and task-general fusion framework based on variations in Centered Kernel Alignment (CKA) similarity across different model layers when trained on diverse datasets. |
| Outcome: | The proposed framework improves performance compared to existing benchmarks. |
Differentiable Instruction Optimization for Cross-Task Generalization (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that instruction tuning is effective for generalizing to arbitrary tasks unseen during training. |
| Approach: | They propose to introduce learnable instructions and optimize them with gradient descent to optimize instruction for generalization ability. |
| Outcome: | The proposed instruction extractor extracts appropriate instruction and improves generalization ability compared to manual instruction tuning. |
Dynamics of Instruction Fine-Tuning for Chinese Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Instruction tuning is a burgeoning method to elicit the general intelligence of Large Language Models. |
| Approach: | They investigate the effects of data quantity, model size, and data construction methods on instruction tuning for Chinese LLMs. |
| Outcome: | The proposed model includes over 40,000 high-quality instruction instances covering ten underlying abilities. |
ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning (2025.findings-emnlp)
Copied to clipboard
Yang Wu, Huayi Zhang, Yizheng Jiao, Lin Ma, Xiaozhong Liu, Jinhong Yu, Dongyu Zhang, Dezhi Yu, Wei Xu
| Challenge: | Prevailing methods for task-specific instruction tuning use similarity metrics to select training data . but instruction tuning loss often fails to exhibit a monotonic relationship with actual task performance . |
| Approach: | They propose a task-specific instruction tuning method that leverages pairwise preference loss as a reward signal. |
| Outcome: | The proposed method surpasses state-of-the-art methods for task-specific instruction tuning. |
CIDAR: Culturally Relevant Instruction Dataset For Arabic (2024.findings-acl)
Copied to clipboard
Zaid Alyafeai, Khalid Almubarak, Ahmed Ashraf, Deema Alnuhait, Saied Alshahrani, Gubran Abdulrahman, Gamil Ahmed, Qais Gawah, Zead Saleh, Mustafa Ghaleb, Yousef Ali, Maged Al-shaibani
| Challenge: | Instruction tuning datasets predominantly cater to English or are derived from English-dominated LLMs. |
| Approach: | They propose to use an Arabic instruction tuning dataset culturally aligned by native Arabic speakers to address drawbacks of finetuning LLMs on machine-generated or machinetranslated datasets. |
| Outcome: | The proposed datasets show that they achieve better cultural alignment than models fine-tuned on other datasets. |
Superfiltering: Weak-to-Strong Data Filtering for Fast Instruction-Tuning (2024.acl-long)
Copied to clipboard
| Challenge: | Earlier studies of instruction tuning on Large Language Models focus on creating large, varied, and high-quality datasets with responses curated by human experts. |
| Approach: | They propose to use a smaller and weaker model to fine tune a larger and stronger model . they find it can largely speed up the data filtering and improve performance . |
| Outcome: | The proposed model can filter instruction data faster and better on benchmarks. |
Federated Data-Efficient Instruction Tuning for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing federated learning (FL) uses all local data, causing excessive computational overhead and overfitting to local data. |
| Approach: | They propose a federated data-efficient instruction tuning approach which utilizes a representative subset of edge-side data to tune LLMs. |
| Outcome: | The proposed method improves Rouge-L on unseen tasks by 10.72% over the SOTA full-data instruction tuning methods while using less than 1.5% of the data samples. |
IT2ACL Learning Easy-to-Hard Instructions via 2-Phase Automated Curriculum Learning for Large Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing studies have focused on the pre-training phase of large language models, but this study focuses on the learning phase of pre-trained LLMs. |
| Approach: | They propose a 2-phase automated curriculum learning guided instruction tuning framework that learns easy-to-hard instructions in a self-adjusting dynamic manner. |
| Outcome: | The proposed framework unlocks latent ability in pre-trained large language models and achieving superior performance across diverse tasks. |
Deep Exploration of Cross-Lingual Zero-Shot Generalization in Instruction Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent studies have focused on instruction tuning to show cross-lingual generalization . a novel non-English meta-dataset is used to study instruction tuning . |
| Approach: | They perform instruction tuning individually for two distinct language meta-datasets and assess the performance on unseen tasks in a non-English language. |
| Outcome: | The proposed model outperforms baseline training in English and Korean by 20.7% and 13.6%. |
Selective Reflection-Tuning: Student-Selected Data Recycling for LLM Instruction-Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Instruction tuning is critical to large language models but its success heavily relies on the training data quality. |
| Approach: | They propose a paradigm that synergizes a teacher LLM’s reflection and introspection with the data selection capability of the student LLM to automatically refine existing instruction-tuning data. |
| Outcome: | The proposed method achieves much stronger and top-tier 7B and 13B LLMs without collecting brand-new data. |
Instruction Matters: A Simple yet Effective Task Selection for Optimized Instruction Tuning of Specific Tasks (2024.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results show that instruction tuning improves zero-shot generalization across various tasks and improves performance of specific tasks. |
| Approach: | They propose a task selection method that leverages instruction information alone to identify relevant tasks and optimize instruction tuning for specific tasks. |
| Outcome: | The proposed method is significantly more efficient than traditional approaches, which require complex measurements of pairwise transferability between tasks or the creation of data samples for the target task. |
Don’t Half-listen: Capturing Key-part Information in Continual Instruction Tuning (2025.acl-long)
Copied to clipboard
Yongquan He, Wenyuan Zhang, Xuancheng Huang, Peng Zhang, Lingxun Meng, Xiang Zhou, Ke Zeng, Xunliang Cai
| Challenge: | Existing methods to improve instruction tuning for large language models may cause catastrophic forgetting (CF) CF is a problem where previously learned abilities are degraded . |
| Approach: | They propose a continual instruction tuning method that uses key-part information gain to replay data and refine training objective. |
| Outcome: | The proposed method achieves superior performance on both seen and held-out tasks. |
From Selection to Refinement: Iterative Optimization for Instruction Data (2026.acl-long)
Copied to clipboard
Hang Hu, Ziyan Liu, Rujie Wen, Ruihui Hou, Xueyan Wu, Mu Zhang, Jianxing Yu, Tong Ruan, Jingping Liu
| Challenge: | Existing methods to optimize instruction tuning datasets face two main challenges: unreasonable pruning of potentially valuable low-quality data and the persistence of noise or semantic drift during revision. |
| Approach: | They propose an automated iterative framework for instruction data optimization that prunes low-quality data and refines low quality data using feedback-driven iteration. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on seven public benchmark datasets with high data efficiency. |