Published October 23, 2026
| Version v1.0.0
Software
Open
Replication package for: Green Code or Costly Code? Evaluating the Energy Costs of Iterative LLM-Based Code Optimizations
Authors/Creators
- 1. University of Augsburg
- 2. King's College London
Description
Replication package for a study of the energy cost of iterative LLM based code optimization. Using the EffiLearner self refinement pipeline, 11 open source models and four decoding configurations were evaluated on all 1000 EffiBench problems, with hardware level energy measurements for both inference and execution. The optimized code is compared with canonical human reference solutions, and break even points are computed for the energy spent on generation. In 42 of 44 configurations the optimized code uses more execution energy than the human reference.
Notes
Files
Sadra-Emami/green_code_or_costly_code-v1.0.0.zip
Additional details
Related works
- Is supplement to
- Software: https://github.com/Sadra-Emami/green_code_or_costly_code/tree/v1.0.0 (URL)
Software
- Repository URL
- https://github.com/Sadra-Emami/green_code_or_costly_code