Papers with Supervised fine-tuning
ATLANTIS: Weak-to-Strong Learning via Importance Sampling (2025.acl-long)
Copied to clipboard
| Challenge: | ATLANTIS is a new technique to improve the performance of large language models. |
| Approach: | They propose a new technique to bridge the gap between the distribution of current datasets and the real-world data distribution by using importance sampling. |
| Outcome: | The proposed technique can bring consistent and significant improvements to models’ performance and can be flexibly transferred among models with different structures. |
Think in Sentences: Explicit Sentence Boundaries Enhance Language Model’s Capabilities (2026.acl-long)
Copied to clipboard
| Challenge: | Existing studies focus on dummy tokens but fail to leverage the inherent sentence-level structure of natural language. |
| Approach: | They propose a method that inserts delimiters at sentence boundaries to enhance large language models' capabilities. |
| Outcome: | The proposed method improves performance on 7B LLMs to 600B Deepseek-V3 with 7.7% gains on GSM8k and 12.5% on DROP. |
LoRAMoE: Alleviating World Knowledge Forgetting in Large Language Models via MoE-Style Plugin (2024.acl-long)
Copied to clipboard
Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Experimental results show that, as the instruction data increases, LoRAMoE can significantly improve the ability to process downstream tasks, while maintaining the world knowledge stored in the LLM. |
| Approach: | They propose a framework that introduces several low-rank adapters and integrates them by using a router network to freeze the backbone model and force a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks. |
| Outcome: | The proposed framework freezes the backbone model and forces a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks, to alleviate world knowledge forgetting. |
How to Fine-Tune Safely on a Budget: Model Adaptation Using Minimal Resources (2025.emnlp-industry)
Copied to clipboard
Anh C. Pham, Mihir Thalanki, Michael Sun, Aditya Chaloo, Ankita Gupta, Tian Xia, Aditya Mate, Ehi Nosakhare, Soundararajan Srinivasan
| Challenge: | Existing methods for fine-tuning safety examples are underdeveloped. |
| Approach: | They hypothesize that the effectiveness of a safety example is governed by its instruction-response behavior and its semantic diversity across harm categories. |
| Outcome: | The proposed method reduces harmfulness by up to 41% while adding only 0.05% more data to the fine-tuning set. |
Rethinking Data Selection at Scale: Random Selection is Almost All You Need (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing data selection techniques are designed for small data pools, a study finds . filtering data by token length is an efficient method for improving results . |
| Approach: | They use self-scoring methods that do not rely on external help to perform fine-tuning . they also find that filtering data by token length offers a stable and efficient method . |
| Outcome: | The proposed methods outperform random selection on large datasets on large data pools. |
Semi-supervised Fine-tuning for Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing LLMs require labeled data, which can be costly in real-world applications. |
| Approach: | They propose a framework that can fully exploit labeled and unlabeled data for LLM fine-tuning . they conducted experiments using GPT-4o-mini and Llama-3.1 on seven general or domain-specific datasets . |
| Outcome: | The proposed framework can fully exploit labeled and unlabeled data for LLM alignment from a propagate-and-select manner. |
Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction (2024.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models may not follow human instructions and produce toxic, hallucinated, or biased content. |
| Approach: | They propose a disperse-then-merge framework that dispersers instruction-following data into portions and trains multiple sub-models using different data portions. |
| Outcome: | The proposed framework outperforms data curation and training regularization on standard knowledge and reasoning benchmarks. |
ToolFlow: Boosting LLM Tool-Calling Through Natural and Coherent Dialogue Synthesis (2025.naacl-long)
Copied to clipboard
Zezhong Wang, Xingshan Zeng, Weiwen Liu, Liangyou Li, Yasheng Wang, Lifeng Shang, Xin Jiang, Qun Liu, Kam-Fai Wong
| Challenge: | Large Language Models (LLMs) can be enhanced by using supervised fine-tuning . however, access to fine-timing data can be limited. |
| Approach: | They propose a Graph-based Sampling strategy and a Planned-generation strategy to enhance the coherence between dialogues by using 8,000 synthetic dialogues. |
| Outcome: | The proposed model achieves tool-calling performance comparable to or surpassing GPT-4 while maintaining strong general capabilities. |
Reinforcement Learning with Supervised Alignment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Supervised fine-tuning (SFT) is a widely used method for adapting Large Language Models to specific tasks. |
| Approach: | They propose a method that uses supervised fine-tuning to train a reward model for reinforcement learning. |
| Outcome: | The proposed method outperforms existing methods on in-domain benchmarks but surpasses them 50 times on out-of-domain and cross-task evaluations. |
InstructDiff: Domain-Adaptive Data Selection via Contrastive Entropy for Efficient LLM Fine-Tuning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing data selection methods suffer from severe domain specificity . existing methods for general instruction-following fail on reasoning tasks . |
| Approach: | They propose a framework that operationalizes contrastive entropy as a domain-adaptive selection criterion through warmup calibration, bi-directional NLL filtering, and entropic-based ranking. |
| Outcome: | Experiments show that InstructDiff outperforms baseline training on reasoning tasks while using only 10% of the data. |
Selecting and Merging: Towards Adaptable and Scalable Named Entity Recognition with Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to align large language models with information extraction tasks are costly and not all training data benefits target domains. |
| Approach: | They propose a framework which dynamically Selects and Merges expert models at inference time and combines experts beneficial to target domains. |
| Outcome: | The proposed framework outperforms the unified model by 10% on multiple benchmarks. |
Not All Parameters Are Created Equal: Smart Isolation Boosts Fine-Tuning Performance (2025.emnlp-main)
Copied to clipboard
| Challenge: | Extensive experiments demonstrate that our approach significantly alleviates task interference and forgetting. |
| Approach: | They propose a framework for supervised fine-tuning for large language models . they first fine-tail the model on each task to identify its core parameter regions . |
| Outcome: | The proposed framework outperforms vanilla fine-tuning and baselines on multiple public benchmarks on reasoning, dialogue, instruction following, and more. |
Data-scarce Behavior Editing of Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Prior studies show that noisy neural circuitries coexist with generalizable abilities within LLMs. |
| Approach: | a new method is proposed to improve the generalizability of large-scale web-based text models . a TaRot method is based on learnable rotation matrices optimized for Bayesian optimization . |
| Outcome: | a new method for task adaptation improves on multiple classification and generation tasks . it improves upon zero- and few-shot performance, with average improvements of 14% and 15% . |
META-LORA: Memory-Efficient Sample Reweighting for Fine-Tuning Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Supervised fine-tuning (SFT) is widely adopted for tailoring large language models (LLMs) to specific downstream tasks. |
| Approach: | They propose a memory-efficient method for automatic sample reweighting that learns to re-weight fine-tuning samples by minimizing the loss on a small, high-quality validation set. |
| Outcome: | Meta-LoRA learns to reweight fine-tuning samples by minimizing the loss on a small, high-quality validation set through an end-to-end bi-level optimization framework based on meta-learning. |
KNN-Instruct: Automatic Instruction Construction with K Nearest Neighbor Deduction (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for generating synthetic instructions for large language models suffer from stale distribution and scalability. |
| Approach: | They propose a method which incorporates KNN deduction to produce meaningful new instructions by summarizing and learning from existing ones. |
| Outcome: | The proposed method outperforms all 7B models on the LMSYS leaderboard. |
ChartM3: A Multi-Stage Code-Driven Pipeline for Constructing Multi-Dimensional and Multi-Step Visual Reasoning Data in Chart Comprehension (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Currently, research on complex chart understanding tasks is limited . a pipeline for visual reasoning datasets addresses these limitations . |
| Approach: | They propose a code-driven pipeline for generating visual reasoning datasets . pipeline integrates retrieval-augmented generation to retrieve professional chart templates . |
| Outcome: | The proposed pipeline enhances chart diversity and data quality through model-based evaluation. |
ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Traditional fine-tuning ignores one-to-many nature of language, leading to overfitting . authors propose a method to fine- tune LLMs by leveraging tokens. |
| Approach: | They propose a method to fine-tune Large Language Models by leveraging tokens to mask low-probability tokens. |
| Outcome: | The proposed method outperforms baselines on general reasoning and mathematical benchmarks. |
Learn Beyond The Answer: Training Language Models with Reflection for Mathematical Reasoning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies focus on *broadening* the training set with data augmentation techniques to maximize such benefits. |
| Approach: | They propose a method that embeds problem reflection into each training instance. |
| Outcome: | The proposed method enhances performance in standard and complex scenarios that require reflective thinking. |
Mitigating Training Imbalance in LLM Fine-Tuning via Selective Parameter Merging (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies suggest that the order of training samples can affect model performance, but this is not the case. |
| Approach: | They propose to merge supervised fine-tuning models with different data orders to mitigate this imbalance by parameter merging. |
| Outcome: | The proposed method outperforms the weighted-average method on five datasets. |
Error Comparison Optimization for Large Language Models on Aspect-Based Sentiment Analysis (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for aspect-based sentiment analysis (ABSA) only compare current predictions and labels on each sample, yet fail to perceive and understand its error outputs from different degrees. |
| Approach: | They propose a framework that can perceive and understand the degree of errors by learning from comparative error pairs. |
| Outcome: | The proposed framework exceeds baselines and achieves the desired performance. |
SAME: Signer-Aware Mixture-of-Experts for Test-Time Adaptation in Sign Language Translation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for supervised fine-tuning are limited due to labeled data . existing methods require long adaptation times and batch statistics are unavailable in streaming settings . |
| Approach: | They propose a plug-and-play, signer-aware Mixture-of-Experts (MoE) TTA architecture for SLT . they use a combination of lightweight MoE modules and unsupervised regularizers to decouple domain shift . |
| Outcome: | The proposed test-time adaptation outperforms existing TTA methods in sign language translation . the proposed architecture can be used in real-world deployments without labeling . |
Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have greatly improved natural language understanding and generation. |
| Approach: | They train a wide range of base models on a variety of datasets including code generation, mathematical reasoning, and general-domain tasks. |
| Outcome: | The results show that training–task synergies persist across all models while others vary substantially, emphasizing the importance of model-specific strategies. |
KaFT: Knowledge-aware Fine-tuning for Boosting LLMs’ Domain-specific Question-Answering Performance (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent literature reveals that supervised fine-tuning (SFT) is suboptimal for domain-specific question-answering tasks. |
| Approach: | They propose a query diversification strategy for robust conflict detection and a knowledge-aware fine-tuning approach to effectively boost LLMs’ performance. |
| Outcome: | The proposed approach improves the model generalization and alleviates the hallucination. |
VCORE: Variance-Controlled Optimization-based Reweighting for Chain-of-Thought Supervision (2026.acl-long)
Copied to clipboard
| Challenge: | Empirical evaluations demonstrate that VCORE achieves the strongest overall average performance, with especially clear gains on lower-capacity models. |
| Approach: | They propose a framework that reformulates supervision as a constrained optimization problem. |
| Outcome: | Empirical evaluations show that VCORE achieves the strongest overall average performance, with especially clear gains on lower-capacity models. |
Learning from Mistakes: Negative Reasoning Samples Enhance Out-of-Domain Generalization (2026.acl-long)
Copied to clipboard
Tian Xueyun, MingHua Ma, Bingbing Xu, Nuoyan Lyu, Wei Li, Heng Dong, Zheng Chu, Yuanzhuo Wang, Huawei Shen
| Challenge: | Recent studies show that supervised fine-tuning (SFT) is a common approach for reasoning in large language models. |
| Approach: | They propose to use supervised fine-tuning (SFT) on chain-of-thought trajectories demonstrations . they find that incorporating negative traxories yields substantial OOD generalization gains . |
| Outcome: | The proposed scheme yields 5.51% OOD gain over positive-only training. |
Explicit Learning and the LLM in Machine Translation (2025.emnlp-main)
Copied to clipboard
| Challenge: | a growing number of researchers are examining whether large language models can learn to translate a "new" language using grammar books. |
| Approach: | They examine an LLM's ability to learn new languages using grammar books . authors suggest alternative fine-tuning strategies to improve explicit learning . |
| Outcome: | The proposed model can learn low-resource languages described in grammar books but lacking extensive corpora. |