DRIFT: Refining Instruction Data via On-Policy Data Attribution

• URL: http://arxiv.org/abs/2606.18307v1
• Date: Tue, 16 Jun 2026 07:21:49 GMT
• Title: DRIFT: Refining Instruction Data via On-Policy Data Attribution
• Abstract summary: Supervised Fine-Tuning (SFT) dictates the capability of Large Language Models (LLMs)<n>Existing data curation methods excel at accelerating training under constrained budgets, but are less suited to elevating the capability upper bound.<n>We propose DRIFT to refine the data distribution toward instances most capable of improving the final model.
• Score: 24.426971578185633
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Optimizing the training data distribution for Supervised Fine-Tuning (SFT) dictates the capability of Large Language Models (LLMs). While existing data curation methods excel at accelerating training under constrained budgets, they are less suited to elevating the capability upper bound. The challenge here is no longer to identify a smaller subset that preserves performance, but to refine the data distribution toward instances most capable of improving the final model. To address this problem, we explore instance-level data attribution using Influence Functions (IF). We identify that standard IF formulations struggle in this setting due to two structural limitations: a proximity gap caused by off-policy validation targets, and a severe bias towards gradient norm. We propose DRIFT (Data Refinement via On-Policy Influence Functions for Supervised Fine-Tuning). Instead of relying on external reference data, DRIFT utilizes the model's on-policy rollouts as validation targets, which empirically minimizes the parameter proximity gap and better aligns with the local neighborhood assumption of IF. It further applies signed weighting based on trajectory correctness and debiases influence scores against the gradient hacking issue, allowing a small set of validation queries to act as reliable anchors for attributing the full dataset. Experiments on 7B-parameter instruction and reasoning models show that DRIFT consistently raises the performance ceiling on both, outperforming existing data curation baselines.

Related papers

• Preference Data Selection for Mitigating the Alignment Tax in Large Language Models [52.18310687523229]
  BALIGN is a balanced data selection strategy that mitigates catastrophic forgetting while optimizing alignment efficacy.<n>By aggregating three key data-centric features into a unified composite risk score, BALIGN systematically filters out high-risk preference samples.<n>Experiments on standard human preference datasets demonstrate that BALIGN strongly preserves foundational capabilities without compromising alignment gains.
  arXiv  Detail & Related papers  (2026-08-25T07:58:42Z)
• Benchmarking Few-shot Transferability of Pre-trained Models with Improved Evaluation Protocols [123.73663884421272]
  Few-shot transfer has been revolutionized by stronger pre-trained models and improved adaptation algorithms.<n>We establish FEWTRANS, a comprehensive benchmark containing 10 diverse datasets.<n>By releasing FEWTRANS, we aim to provide a rigorous "ruler" to streamline reproducible advances in few-shot transfer learning research.
  arXiv  Detail & Related papers  (2026-02-28T05:41:57Z)
• Can Small Training Runs Reliably Guide Data Curation? Rethinking Proxy-Model Practice [109.9635246405237]
  We show that the experiment conclusions about data quality can flip with even minor adjustments to training hyper parameters.<n>We introduce a simple patch to the evaluation protocol: using reduced learning rates for proxy model training.<n> Empirically, we validate this approach across 23 data recipes covering four critical dimensions of data curation.
  arXiv  Detail & Related papers  (2025-12-30T23:02:44Z)
• Data Efficient Adaptation in Large Language Models via Continuous Low-Rank Fine-Tuning [34.343514432589586]
  This paper proposes textbf, a novel framework that integrates Low-Rank Adaptation (LoRA) with a continuous fine-tuning strategy.<n> Experiments on 15 diverse datasets show that DEAL consistently outperforms baseline methods.<n>These findings demonstrate the potential of our approach to advance continual adaptation in Large Language Models.
  arXiv  Detail & Related papers  (2025-09-23T12:55:57Z)
• Mind the Gap: Data Rewriting for Stable Off-Policy Supervised Fine-Tuning [33.899779762210976]
  Supervised fine-tuning (SFT) of large language models can be viewed as an off-policy learning problem.<n>Existing methods mitigate this issue with KL penalties or clipping, which passively updates rather than actively reducing the gap.<n>We propose a simple yet effective data rewriting framework that proactively shrinks the policy gap before training.
  arXiv  Detail & Related papers  (2025-09-18T17:02:30Z)
• Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment [16.352863226512984]
  Test-time adaptation (TTA) enhances the zero-shot robustness under distribution shifts by leveraging unlabeled test data during inference.<n>Most methods rely on backpropagation or iterative optimization, which limits scalability and hinders real-time deployment.<n>We propose ADAPT, an Advanced Distribution-Aware and back propagation-free Test-time adaptation method.
  arXiv  Detail & Related papers  (2025-08-21T13:42:49Z)
• DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with Limited Data [7.70699448711673]
  Decision-focused learning (DFL) offers an end-to-end approach to the predict-then-optimize (PO) framework by training predictive models directly on decision loss (DL)<n>Some predictive models are non-differentiable or black-box, which cannot be adjusted using gradient-based methods.<n>We propose a novel framework, Decision-Focused Fine-tuning (DFF), which embeds the DFL module into the PO pipeline via a novel bias correction module.
  arXiv  Detail & Related papers  (2025-01-03T15:46:25Z)
• Reward-Augmented Data Enhances Direct Preference Alignment of LLMs [63.32585910975191]
  We introduce reward-conditioned Large Language Models (LLMs) that learn from the entire spectrum of response quality within the dataset.<n>We show that our approach consistently boosts DPO by a considerable margin.<n>Our method not only maximizes the utility of preference data but also mitigates the issue of unlearning, demonstrating its broad effectiveness beyond mere data expansion.
  arXiv  Detail & Related papers  (2024-10-10T16:01:51Z)
• ASFT: Aligned Supervised Fine-Tuning through Absolute Likelihood [14.512464277772194]
  Aligned Supervised Fine-Tuning (ASFT) is an effective approach that better aligns Large Language Models with pair-wise datasets. ASFT mitigates the issue where the DPO loss function decreases the probability of generating human-dispreferred data. Extensive experiments demonstrate that ASFT is an effective alignment approach, consistently outperforming existing methods.
  arXiv  Detail & Related papers  (2024-09-14T11:39:13Z)
• REST: Enhancing Group Robustness in DNNs through Reweighted Sparse Training [49.581884130880944]
  Deep neural network (DNN) has been proven effective in various domains. However, they often struggle to perform well on certain minority groups during inference.
  arXiv  Detail & Related papers  (2023-12-05T16:27:54Z)
• Better Practices for Domain Adaptation [62.70267990659201]
  Domain adaptation (DA) aims to provide frameworks for adapting models to deployment data without using labels. Unclear validation protocol for DA has led to bad practices in the literature. We show challenges across all three branches of domain adaptation methodology.
  arXiv  Detail & Related papers  (2023-09-07T17:44:18Z)
• CAFA: Class-Aware Feature Alignment for Test-Time Adaptation [50.26963784271912]
  Test-time adaptation (TTA) aims to address this challenge by adapting a model to unlabeled data at test time. We propose a simple yet effective feature alignment loss, termed as Class-Aware Feature Alignment (CAFA), which simultaneously encourages a model to learn target representations in a class-discriminative manner.
  arXiv  Detail & Related papers  (2022-06-01T03:02:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.