Kuangrepi的博客

 
//////////////////////////////tt///////////////////ttt//t////////||/////////////
//////////////////////ttttt////tt///////////////////t///////////////////////////
/tt/tt/////////////ttttttttttttt//////////////////////t/////////////////////////
//ttttt/////t///////tt//t////////////////ttt/t//////////////////////////////////
tt//////ttttt///////////////////////////tttt//ttttt//////////ttt///////||||/////
ttt///////////////////////////////////tttttt/tttttttttt////////////////|||//////
/t///////////////////t////////t////tt///ttt////tttttttt//////////////////||/////
/////////////t////////////////////tt////////////////t///////////////////////////
///t/tttt///////////////////////////////////////////////////////////////////////
//////ttt//////////////////tt///////////////////ttt/////////////////////////////
///////tttt////////////////////ttttttttt/t//////////////////////////////////////
///////ttt//////////////////tttttttt/1{]--}{|///////////////////////////////////
////////////////////////////)]_I^I:!''. ..''```^<1/tt///////////////////////////
///////////////////////t(_;`'"'.''.'.. . .   ..``^";[t/////////////////t////////
/////////////////////t):.``  .`.      .    ... .'`^``,>|////////////////////////
////////////////////|!.'.'' .            .  ... .'''`'`;+///////////////////////
///////////////////|:'''''...  .'''...........'.`..'"^"":i1/////////////////////
///////tt/////////]''''..'''.`"li>!!`'","",^`''`'^``,"",":![t//////////////////t
////////tttttt//t+''......^`:l<??1}}>l??{]~-?!Il;,,l",:!',ll_t///////////////ttt
///////tttt/////[.`'..' ''`:~+[)|tff/]xxnvcvnzfxr}li:,^;".^!l]//////////////////
////////t//////ti.`'....':il?(/jrruvczXUJCQQOmwmqmQx}>"^I'.;il|/////////////////
///////tt/t///t|^'''...'I-{(trnvcXYJCLLQ00Q0Zqdbkbbbpwti>;;";~1ttttt//tt////////
////////ttttttt/+l'. .'l[(|t//juvYYUJJCL0QQ0Ompppqqwpdp0|>:^;i}tttt////t////////
//////////tttttt/+''.'l}(1~>~--~++++]|nzJLJj1[?}{]zvYqQ|I:,:]tttt////tt///////
//////////ttttttt/"`'"{//[~<<_[11[++}})rULUuf]_?{|ff|/vZp0>l;~/tt/t///ttt///////
/////////////ttttt_'.:;.,t(1}-_]{?cr|("',;!Ur)}111xQmwqJ;!<II/tttt///////t//////
/|///////t/tttttttf],!?)<t(-<[1<[f-||[,|v|?Cvf[/_?LUrc0rUL>/fjt/t///////////////
/||//////tttttttt(){_Irj/ftttfrrjrjf(IlnZk+t0QUzzJOwqpdzbb1LOuxJ////////////////
|||////////ttttt)>-)-"rrr(jnuuvvunjt]<(vOpw|Xwm0QQOwdqxbbdrCq0)v/tt/////////////
||/////////tttttt}]|?I|rrrrrnxxxjtt(}}jzLwpwLOO0LQQmdhahbqtnmCXrttt/////////////
///////////////ttt{{<--jjrrxuvccvx/](|jcLwbdpOQOmwqbkhhhbqrnw0f/////////////////
///////|/////////tt1{[{(jjjruvzXcx(_j(-(xUf/ZCULOmpkkkkkdmZC0t//////////////////
/////////////////ttt/t|{rrjrxuczcxjt(11)fXZmwwZOOmpkhhhkppdY////tt//t////t//////
////////////////tttttt/(trrjxnuuunxjjfjXXXmqpdwwqqpbkkkbY////t/ttttttt//t///////
////////////////tttttttttrrjrxxnuut1]]??(vuuzOznOpdbbkkdt///tt//ttttttt/////////
/////////////////tttttttt/jjjjrrxf}--}trnX0ZLZCjXZdbbbdx/ttttttt//ttttttt///////
////ttt////////////ttttttt/fttffftt/|/jxucCZwqwm0Owddqrttttt////tttttttttttt////
/////////////////////tttttf(((|////tfxcYL0mqdpppqwwmwf//tttttt/tttt/tttttt//////
/////////////////////tttttf(){})(|/tjxvYLZqbkbppw0Ldbftttttttt/t/ttttttttttt////
/////////////////////tttttf(((1{}}{)|trnzYL00LUXCZpbdftttt/ttt/////t////tttt////
/|////////////////tttttttt/||||(1{{{1)(/tjncULOmqdkbdntttt///t///ttttt////t/////
////////////////tttttttttf|||t/||)1)1)|fuYQOmqpdbbkkbJ///////ttttt//////////////
//////////////////ttttttt||//tt/||||||/fvLZmqdkhkkkbbpJ/////tttttt//////////////
/////////////////tttt{])(|/tttttft////fxXLZqpkhkkbhkbddpztttttt///t/////////////
////////////t///tt);;{)(///tttfjjjft/txcU0wpbkbbbkhkkbbkbdp??tttttt/////////////
////////////t1~:"`''`[|/t/fjjjjjrxxfttjvUQmqqqppdkkkkkhhhhhj~_]|tfttt///////////
/////////1I""^```'''''!|tffjjjrjjxuunrruzCOZZmpppdkhkho*ok}>>><~<]i)tttt////////
/////|~""^^```````'''''`<|tjrxunjruXXcczUL000mpdpdhhkhhbn>l!i!i!!~-++_])tt//////
///+:"^^^^^^"^^^^^``'````''',)nccruJCJUXUJCCQwdpmpkkqnillIIIIIII>~__+>~~>_-tt///
>"""""^^""""""""""""^^``^^^^````^,>(cCLCJJJUUJUzf]>lIII;;;;;;;I<<~~~>+<>Ili<_)//
`"^""",,,,"""""^^"""""""`'^^^^^^^^^^^^^^"""""",::::,;;;;;:;:li<ii>i>><<+>IlII!>[
"^",,,,,,,""""^^^^^^""""""^^`''`^^""""",,"""",",,::;:,,,;I!l!!!!l!i>>~~~~>illIIl
""",,,,,,""""^"""""^""""""""""^^^^^^```^^^^^``^":::;II;;IIIIllllll!<~+__-___~ilI

谷羽Yu Gu

南京大学 智能科学与技术 本科生Undergraduate, Intelligence Science and Technology, Nanjing University

231880100@smail.nju.edu.cn

教育经历Education

南京大学 · 智能科学与技术 · 本科 Nanjing University · Intelligence Science and Technology · Bachelor's Degree 2023.09 – 预计 2027.06Sep 2023 – Expected Jun 2027
江苏省苏州市 · GPA 87.6 / 100 Suzhou · GPA 87.6 / 100

科研 / 实习经历Research & Internship Experience

中国科学院物理研究所 · 凝聚态物质科学数据中心Institute of Physics, Chinese Academy of Sciences · Condensed Matter Science Data Center 2025.09 – 2026.01Sep 2025 – Jan 2026
AI工程师实习生 · 北京 AI Engineer Intern · Beijing
指导教师：蔡云麒；日常工作由博士生郑子扬带领Advisor: Cai Yunqi; day-to-day work led by PhD student Zheng Ziyang

主导 MatSciBench 自动化扩展与开发构建项目，完成智能体框架开发，将数据集自动化搜集、处理与构建及自动模型训练流程打通，促进数据中心构建。 Led the automated extension and development of MatSciBench — built an agent framework and connected automated dataset collection, processing, construction, and model-training pipelines to support the data center.

相关知识库：Related wiki: agent
新加坡国立大学National University of Singapore 2025.11 – 2026.02Nov 2025 – Feb 2026
科研实习 / 推进实验人员 · 新加坡 Research Assistant · Singapore
指导教师：郑执；通讯作者：Wee Sun LeeAdvisor: Zhi Zheng; Corresponding author: Wee Sun Lee

负责 Soft Thinking 方法中推理重叠部分的开发，加快了推理速度；参与论文校对与修改，作为第二作者参与完整科研流程，产出论文 SofT-GRPO。 Developed the inference-overlap component for the Soft-Thinking method to speed up inference; proofread and revised the manuscript as second author, resulting in the paper SofT-GRPO.

相关知识库：Related wiki: rl-llm-training
南方科技大学 · CIAM 实验室Southern University of Science and Technology · CIAM Lab 2026.02 – 至今Feb 2026 – Present
研究人员 · 深圳 Research Assistant · Shenzhen
指导教师：王振坤Advisor: Wang Zhenkun

主要研究 ES（进化策略）算法与模型融合算法，零阶优化（ZO）为次要方向；从事算法设计与改进研究，论文投稿中。 Primarily researching Evolution Strategies (ES) and model merging algorithms, with zeroth-order optimization (ZO) as a secondary direction; working on algorithm design and improvement, with a paper currently under submission.

相关知识库：Related wiki: es-algorithm, muon, model-fusion
华为诺亚方舟实验室Huawei Noah's Ark Lab 2026.07 – 至今Jul 2026 – Present
Agent 算法实习生 · 深圳 Agent Research Intern · Shenzhen

从事 Agent 相关研发工作，与南方科技大学 CIAM 实验室的研究同时进行。 Working on Agent-related research and development, concurrent with ongoing research at SUSTech CIAM Lab.

相关知识库：Related wiki: agent

发表论文Publications

Zhi Zheng, Yu Gu, Wei Liu, Yee Whye Teh, Wee Sun Lee*

SofT-GRPO: Surpassing Discrete-Token LLM Reinforcement Learning via Gumbel-Reparameterized Soft-Thinking Policy Optimization

预印本Preprint

语言 / 考试成绩Language Tests

IELTS 7.0 Writing 6.5 · Reading 8.5 · Listening 6.5 · Speaking 6.0
大学英语六级CET-6 551
0%