Skip to content
Trending storyDeveloping

AutoSFT: multi-agent system for automated SFT data research

1 article1 sourcesince Nov 5Last article 1h ago ·

Overview

AISummary of 1 article

THUNLP, BUPT, ModelBest, UIUC and CUHK present AutoSFT, a multi-agent system for automating supervised fine-tuning (SFT) data research.

It splits the work into an Executor, four Analysis Subagents covering training data, failure cases, learning dynamics and research history, a Synthesis writer with a verifier, and a meta agent that picks the next recipe.

In 30 rounds, AutoSFT uses all six action families and reaches an IFEval score of 63.33 in round 6, above the single-agent baseline's best over 50 rounds. It reaches 67.10 by round 30. The authors say automating data research may depend more on covering diverse data actions than on a stronger single action.

Written by AI from the articles below · updated Nov 5, 9:15 AM ET

Check the sources:

Latest developmentNov 5, 9:01 AM ET
AutoSFT automates data research with specialized agent roles for SFT

Article timeline

The articles in this story. Times are ET.

Nov 5
  1. OpenBMBOfficial
    AutoSFT automates data research with specialized agent roles for SFT

    AITHUNLP, BUPT, ModelBest, UIUC and CUHK present AutoSFT, a multi-agent system that splits SFT data research into an Executor, four Analysis Subagents, a Synthesis writer with a verifier, and a meta agent choosing the next recipe. In 30 rounds, AutoSFT uses all six action families and reaches IFEval 63.33 in round 6, above the single-agent baseline's best over 50 rounds, and 67.10 by round 30. The authors say automating data research may depend more on covering diverse data actions than on a stronger single action.

    Image from @OpenBMB's post

Heat Trend

Not enough continuous observations to show a trend yet.