Auto-Labeling Code Datasets With Sustainability Metrics for LLM Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large Language Models (LLMs) do not consider the environmental impact or 'greenness' of software code solutions during training, necessitating separate evaluation for energy-efficient or environmentally conscious code suggestions.
Innovation Solution
A processing device collects energy metrics of software code executing in a computing environment, determines a sustainability label based on these metrics, and assigns it to the code to create a sustainability-based dataset, which is integrated into the LLM training regimen.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If LLMs are trained with traditional datasets, then they can generate code that meets functional requirements, but they do not consider environmental impact or energy efficiency of the generated code
Solution Approach 1:
The system performs preliminary actions by collecting sustainability information (energy metrics, sustainability scores) for code snippets before they are used to train the LLM. This advance preparation ensures that sustainability data is available and integrated into the training process, allowing the model to learn about environmental impact without compromising functional code generation capabilities
Solution Approach 2:
The system introduces an intermediary component that bridges traditional code datasets and sustainability considerations. This intermediary layer collects, processes, and attaches sustainability metadata to code snippets, enabling the LLM to access environmental impact information during training without disrupting the existing code generation workflow
2Object-affected harmful factors
If sustainability labeling system is implemented, then LLMs can generate environmentally conscious code, but the system complexity increases due to additional data collection and processing requirements
Solution Approach 1:
The sustainability labeling system is designed with multi-functionality to reduce overall system complexity. The same infrastructure used for code collection and management is leveraged to gather sustainability data, and the labeling mechanism integrates with existing dataset formats. This universal approach allows a single system to handle both traditional code processing and sustainability evaluation without requiring completely separate complex subsystems
Solution Approach 2:
The system implements self-service mechanisms where code snippets and their sustainability information are automatically collected and labeled without requiring extensive manual intervention. Automated scripts execute code in controlled environments to gather energy metrics, and algorithms automatically generate sustainability labels based on collected data, reducing the operational complexity of maintaining the sustainability labeling infrastructure
Data Source
AI summary
Systems and methods are disclosed that deploy software code from a dataset into a computing environment. The systems and method collect energy metrics of the software code while executing in the computing environment. The systems and methods determine a sustainability label for the software code based on the energy metrics. The systems and methods assign the sustainability label to the software code to produce a sustainability-based dataset.