본문

서브메뉴

Customizing Large Language Models: From Synthetic Data Generation to Efficient Deployment
Customizing Large Language Models: From Synthetic Data Generation to Efficient Deployment ...
Customizing Large Language Models: From Synthetic Data Generation to Efficient Deployment

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260311091517.5
ISBN  
9798270232726
DDC  
005
저자명  
Wen, Yeming
서명/저자  
Customizing Large Language Models: From Synthetic Data Generation to Efficient Deployment / Yeming Wen
발행사항  
[Sl] : The University of Texas at Austin, 2025
형태사항  
1 electronic resource (135 pages)
주기사항  
Source: Dissertations Abstracts International, Volume: 87-06, Section: B.
주기사항  
Advisors: Chaudhuri, Swarat Committee members: Durrett, Greg; Akella, Aditya; Jermaine, Chris.
학위논문주기  
- Ph.D. : The University of Texas at Austin, 2025.
초록/해제  
요약Large Language Models (LLMs) have demonstrated remarkable capabilities in various tasks. However, generating responses that accurately meet the requirements of execution-based scenarios remains challenging. This challenge stems fundamentally from the pretraining objectives of LLMs, where next-token prediction is not inherently execution-aware, making customization in the post-training stage essential. This thesis explores a comprehensive approach to customizing LLMs for execution-driven contexts through three interconnected works.First, we investigate the potential of execution-derived feedback for enhancing LLM performance in specification-rich tasks. We propose a framework that synthesizes training data by combining natural language intents with specifications derived from execution. These specifications help to ground the model's understanding of task requirements, also addressing the ambiguity often present in natural language prompts. Our evaluations demonstrate substantial improvements in generating solutions that comply with execution specifications in data science notebook and tool automation.As synthetic data generation scales, we observe diminishing returns when fine-tuning a single model. To address this limitation, our second work explores how to make full use of abundant synthetic data by training multiple specialized model variants. We use influence functions to efficiently group synthetic data, promoting diversity in LLM outputs while maintaining quality. Experiments showed that our approach can diversify foundation model responses while maintaining high quality in the code generation domain and several tasks in the natural language understanding domain.Finally, we address the practical challenges of deploying these customized models with fLoRA (Fast Low-Rank Adaptation). fLoRA enables efficient, real-time serving of task-specific model variants through batched low-rank adaptation, making it computationally feasible to maintain and deploy multiple specialized models while ensuring that the most appropriate variant is selected for a given execution-based task. Together, these contributions form a holistic strategy for customizing LLMs in execution-driven scenarios, enhancing both their accuracy, diversity and efficiency across applications in code generation and tool automation. 
언어주기  
English
일반주제명  
Computer engineering
일반주제명  
Computer science
일반주제명  
Information technology
키워드  
Large Language Models
키워드  
Synthetic data
키워드  
Natural language prompts
키워드  
Tool automation
키워드  
Code generation
기타저자  
The University of Texas at Austin Computer Science
기본자료저록  
Dissertations Abstracts International. 87-06B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260311s2025        us                                    eng  d
■001000017361247
■00520260311091517.5
■006m          o    d                
■007cr|nu||||||||
■020    ▼a9798270232726
■040    ▼aMiAaPQD▼beng▼cMiAaPQD▼erda
■082    ▼a005
■1001  ▼aWen,  Yeming▼eauthor.
■24510▼aCustomizing  Large  Language  Models:  From  Synthetic  Data  Generation  to  Efficient  Deployment  ▼cYeming  Wen
■260    ▼a[Sl]▼bThe  University  of  Texas  at  Austin▼c2025
■264  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a1  electronic  resource  (135  pages)
■336    ▼atext▼btxt▼2rdacontent
■337    ▼acomputer▼bc▼2rdamedia
■338    ▼aonline  resource▼bcr▼2rdacarrier
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-06,  Section:  B.
■500    ▼aAdvisors:  Chaudhuri,  Swarat    Committee  members:  Durrett,  Greg;  Akella,  Aditya;  Jermaine,  Chris.
■5021  ▼bPh.D.▼cThe  University  of  Texas  at  Austin▼d2025.
■520    ▼aLarge  Language  Models  (LLMs)  have  demonstrated  remarkable  capabilities  in  various  tasks.  However,  generating  responses  that  accurately  meet  the  requirements  of  execution-based  scenarios  remains  challenging.  This  challenge  stems  fundamentally  from  the  pretraining  objectives  of  LLMs,  where  next-token  prediction  is  not  inherently  execution-aware,  making  customization  in  the  post-training  stage  essential.  This  thesis  explores  a  comprehensive  approach  to  customizing  LLMs  for  execution-driven  contexts  through  three  interconnected  works.First,  we  investigate  the  potential  of  execution-derived  feedback  for  enhancing  LLM  performance  in  specification-rich  tasks.  We  propose  a  framework  that  synthesizes  training  data  by  combining  natural  language  intents  with  specifications  derived  from  execution.  These  specifications  help  to  ground  the  model's  understanding  of  task  requirements,  also  addressing  the  ambiguity  often  present  in  natural  language  prompts.  Our  evaluations  demonstrate  substantial  improvements  in  generating  solutions  that  comply  with  execution  specifications  in  data  science  notebook  and  tool  automation.As  synthetic  data  generation  scales,  we  observe  diminishing  returns  when  fine-tuning  a  single  model.  To  address  this  limitation,  our  second  work  explores  how  to  make  full  use  of  abundant  synthetic  data  by  training  multiple  specialized  model  variants.  We  use  influence  functions  to  efficiently  group  synthetic  data,  promoting  diversity  in  LLM  outputs  while  maintaining  quality.  Experiments  showed  that  our  approach  can  diversify  foundation  model  responses  while  maintaining  high  quality  in  the  code  generation  domain  and  several  tasks  in  the  natural  language  understanding  domain.Finally,  we  address  the  practical  challenges  of  deploying  these  customized  models  with  fLoRA  (Fast  Low-Rank  Adaptation).  fLoRA  enables  efficient,  real-time  serving  of  task-specific  model  variants  through  batched  low-rank  adaptation,  making  it  computationally  feasible  to  maintain  and  deploy  multiple  specialized  models  while  ensuring  that  the  most  appropriate  variant  is  selected  for  a  given  execution-based  task.  Together,  these  contributions  form  a  holistic  strategy  for  customizing  LLMs  in  execution-driven  scenarios,  enhancing  both  their  accuracy,  diversity  and  efficiency  across  applications  in  code  generation  and  tool  automation. 
■546    ▼aEnglish
■590    ▼aSchool  code:  0227
■650  4▼aComputer  engineering
■650  4▼aComputer  science
■650  4▼aInformation  technology
■653    ▼aLarge  Language  Models
■653    ▼aSynthetic  data
■653    ▼aNatural  language  prompts
■653    ▼aTool  automation
■653    ▼aCode  generation
■7102  ▼aThe  University  of  Texas  at  Austin▼bComputer  Science.▼edegree  granting  institution.
■7201  ▼aChaudhuri,  Swarat▼edegree  supervisor.
■7730  ▼tDissertations  Abstracts  International▼g87-06B.
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17361247▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    detalle info

    • Reserva
    • No existe
    • Mi carpeta
    • Primera solicitud
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    Material
    número de libro número de llamada Ubicación estado Prestar info
    TF18081 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Las reservas están disponibles en el libro de préstamos. Para hacer reservaciones, haga clic en el botón de reserva

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.