Arxiv 2609 24432v1 1 Of Tokens Can Be Enough On Gradient Estimation I

--
name: arxiv-2609-24432v1-1-of-tokens-can-be-enough-on-gradient-estimation-i
description: '1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation (arXiv: 2609.24432v1)'
metadata:
  {
    "arxiv_id": "2609.24432v1",
    "utility": 1.0,
    "title": "1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation",
    "authors": "Huanxin Sheng, Zhiling Ye, Haonan Wang, Jian Wang, Jinjie Gu, Jian Kang",
    "url": "http://arxiv.org/abs/2609.24432v1"
  }
--

# 1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation

**arXiv ID:** 2609.24432v1
**Authors:** Huanxin Sheng, Zhiling Ye, Haonan Wang, Jian Wang, Jinjie Gu, Jian Kang
**URL:** http://arxiv.org/abs/2609.24432v1
**Utility Score:** 1.00

## Summary

This skill was automatically generated from the arXiv paper titled "1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation" (ID: 2609.24432v1).

## Usage

This skill can be used to reference the paper's concepts, methodologies, or findings in agent workflows.

## References

- arXiv: http://arxiv.org/abs/2609.24432v1