Length Instruction Fine-Tuning with Chain-of-Thought (LIFT-COT): Enhancing Length Control and Reasoning in Edge-Deployed Large Language Models
Open Access
- 19 April 2025
- journal article
- research article
- Published by MDPI AG in Electronics
- Vol. 14 (8) , 1662
Abstract
This paper investigates the effectiveness of Chain-of-Thought (COT) reasoning in addressing length bias when deploying Large Language Models (LLMs) on the device side within wireless network environments. By conducting systematic experiments on two representative datasets, we perform a comparative evaluation of LLM performance under standard prompting versus COT-augmented prompting, using two benchmark tests. The study emphasizes how the COT strategy regulates response length and improves compliance with predefined maximum length (max_len) constraints. Experimental results demonstrate that COT reasoning significantly mitigates length bias, enhances output length control, and improves overall performance. These findings contribute to the theoretical foundation of LLM deployment in edge computing scenarios and provide actionable insights for future research on resource-constrained language model applications.Keywords
Funding Information
- Technical Foundation of the Equipment Development Department (JZX7J202104ZL002600)
This publication has 18 references indexed in Scilit:
- E-GAIL: efficient GAIL through including negative corruption and long-term rewards for robotic manipulationsApplied Intelligence, 2025
- Enhancing Policy Generation with GraphRAG and YouTube Data: A Logistics Case StudyElectronics, 2025
- Hallucination Detection in Foundation Models for Decision-Making: A Flexible Definition and Review of the State of the ArtACM Computing Surveys, 2025
- An Innovative Solution to Design Problems: Applying the Chain-of-Thought Technique to Integrate LLM-Based Agents With Concept Generation MethodsIEEE Access, 2024
- Hallucination Reduction and Optimization for Large Language Model-Based Autonomous DrivingSymmetry, 2024
- ERG-AI: enhancing occupational ergonomics with uncertainty-aware ML and LLM feedbackApplied Intelligence, 2024
- Zero and Few Short Learning Using Large Language Models for De-Identification of Medical RecordsIEEE Access, 2024
- Enhancing Zero-Shot Crypto Sentiment With Fine-Tuned Language Model and Prompt EngineeringIEEE Access, 2024
- Enhancing yes/no question answering with weak supervision via extractive question answeringApplied Intelligence, 2023
- Continuum and thermodynamic limits for a simple random-exchange modelStochastic Processes and their Applications, 2022