AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning
Summary
AutoSFT 提出一个多智能体系统，将 SFT 数据研究中的执行、分析与合成分离，自动执行多样化的数据操作，相比单智能体在 IFEval、IFBench 和 MBPP+ 上取得显著提升。研究发现自动化 SFT 数据研究的关键在于覆盖多样化的数据操作，而非强化单一操作。
View Cached Full Text
Cached at: 10/16/26, 03:01 AM
Paper page - AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning
Source: https://huggingface.co/papers/2610.22194 Authors:
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
Abstract
AutoSFT uses multi-agent collaboration to automate diverse supervised fine-tuning data actions, improving evaluation scores over single-agent systems by separating execution, analysis, and synthesis.
Language model agents increasingly automate post-training, but when given the full training stack they converge onsupervised fine-tuningand spend most effort revising training data. Existing systems give the agent one kind of control, selecting rows from a fixed pool or calling a generator, whereas a human data engineer adjusts proportions, prunes, repairs, replaces and synthesizes the data. We design adata action spacefor these changes and ask whether an automated system can make the full range rather than one of them, and what structure that requires. We first run this loop with a single agent, and it reaches only one action, adjusting category proportions, while its context fills with execution output and is discarded mid-search. We therefore introduce AutoSFT, amulti-agent systemwhere ameta agentseparates execution from analysis, gathers proposals from independent angles, and calls asynthesis subagentfor capabilities the pool cannot cover, with the model, the trainer, and the budget fixed within each comparison. Adding each component in turn helps: the executor halves context loss, the analysis subagents lift the bestIFEvalby +6.5, and synthesis adds +3.4IFEvaland +2.9IFBench. The same loop transfers to code, where analysis adds +3.7MBPP+over a single agent that does not improve. Every result comes from runs that retrain from the base model each round, so each component’s effect is individually attributable. Our findings suggest that automating SFT data research depends more on covering diverse data actions than on a stronger single action.
View arXiv pageView PDFGitHubAdd to collection
Get this paper in your agent:
hf papers read 2610\.22194
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.22194 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.22194 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.22194 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
@aakaran31: SFT is not dead! We found a way to make SFT rival current prevailing posttraining methods, often generalizing better an…
Authors argue supervised fine-tuning (SFT) is not obsolete, presenting a method that brings SFT on par with prevailing posttraining approaches like RL and OPSD, often generalizing better and forgetting less, by introducing sampling into the posttraining stack.
ProCUA-SFT Technical Report
ProCUA-SFT is a large-scale synthetic dataset of 3.1M step-level SFT samples for training computer-use agents, produced via an automated pipeline using a single VLM (Kimi-K2.5). Fine-tuning UI-TARS 7B on it achieves 45.0% on OSWorld, an 18.7 point improvement over the base model.
Shared SFT Lessons Across Alignment, Model Organisms, and Toy Models
This paper studies transferring lessons about supervised fine-tuning (SFT) across alignment training, model organisms, and toy models, showing that techniques like training on reasons for behavior and mixing on-model data can improve generalization and capability preservation.
SFT or RL for Tool-Calling Agents? A Controlled Study Across Data, Method, and Scale
This controlled study compares supervised fine-tuning and reinforcement learning methods for training tool-calling agents across different datasets and model scales, finding that SFT with LoRA is strongest in-distribution while RL shows slight advantages in cross-dataset transfer.
Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning
AuroSFT introduces a parameter-efficient adapter-based framework for multi-task supervised fine-tuning that rolls back adapter checkpoints at task-wise peaks instead of full-model checkpoints, achieving higher average accuracy than mSFT.