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. |
Similar Papers
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. |
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. |
Phased Instruction Fine-Tuning for Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to enhance pre-trained language models' ability to follow instructions are limited due to the simultaneous handling of varying instruction complexities. |
| Approach: | They propose a phased instruction fine-tuning method that posits that the transition of a pre-trained language model from simple next-word prediction to sophisticated instruction following is a gradual learning process. |
| Outcome: | The proposed method surpasses the one-off instruction fine-tuning method in win rate and validates the hypothesis of progressive alignment. |
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. |
The Emperor’s New Reasoning: Format Imitation Overshadows Genuine Mathematical Understanding in SFT (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have yielded impressive gains on mathematical reasoning benchmarks via supervised fine-tuning (SFT). |
| Approach: | They investigate the mechanisms behind SFT improvements in small-scale large language models by examining four key questions: (1) Are performance gains primarily due to format alignment rather than reasoning? (2) Can high-quality supervision encourage genuine reasoning? (4) Are format alignment gains consistent across model sizes and architectures? |
| Outcome: | The proposed models outperform the proprietary models on OlympiadBench and Omni-Math, but lack the brittleness of the models under perturbations to test their reasoning abilities. |
A Layer-wise Analysis of Supervised Fine-Tuning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning ignore depth-dependent heterogeneity of instruction-following . a critical gap remains in understanding where these changes occur across the model's depth and which layers are essential for instruction- following. |
| Approach: | They propose a method which selectively updates critical intermediate layers . they show that effective alignment is architecturally localized rather than distributed . |
| Outcome: | The proposed method outperforms standard LoRA up to 10.2% on GSM8K with reduced parameter overhead. |
SFTMix: Elevating Language Model Instruction Tuning with Mixup Recipe (2026.acl-long)
Copied to clipboard
| Challenge: | Efforts to improve instruction tuning often focus on higher-quality supervised fine-tuning datasets, typically requiring data filtering with proprietary LLMs or human annotation. |
| Approach: | They propose a Mixup-based recipe that elevates LLM instruction tuning without relying on well-curated datasets. |
| Outcome: | The proposed model improves instruction-following and healthcare-specific tasks with consistent improvements across LLM families and SFT datasets. |
A Deep Dive into the Trade-Offs of Parameter-Efficient Preference Alignment Techniques (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models are pre-trained on trillions of tokens and instruction-tuned or aligned to specific preferences. |
| Approach: | They propose guidelines to help researchers perform more effective parameter-efficient LLM alignment. |
| Outcome: | The proposed methods outperform preference optimization and outperformed pre-trained models on three key axes. |
From Distributional to Overton Pluralism: Investigating Large Language Model Alignment (2025.naacl-long)
Copied to clipboard
| Challenge: | a large language model's (LLM) output distribution is changed by an alignment process . a recent study shows that aligned models surface information that cannot be recovered from base models without fine-tuning. |
| Approach: | They analyze two aspects of the alignment process that change output distributions . they find alignment suppresses irrelevant and unhelpful content . |
| Outcome: | The proposed model can be imitated without fine-tuning by using in-context examples and lower-resolution semantic hints about response content. |
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. |