Infrastructure scenarioBaselineDatacenter PUE1.20Grid intensity0.350 kg CO₂e/kWh
Estimate one training run
Training compute4.20e+22 FLOPsIT energy77,777.8 kWhFacility energy93,333.3 kWhOperational CO₂e32,666.7 kg
My roleIndependent engineerStackPython · Streamlit · Pandas · Plotly
Problem
Energy and carbon estimates depend on infrastructure assumptions.
Results change based on model size, training tokens, hardware efficiency, data-center overhead, and grid intensity. EcoMind displays those inputs and allows users to change them before comparing results.
Implementation
Model inputs are converted into compute, energy, and carbon estimates.
The application converts parameters and training tokens into FLOPs, IT energy, facility energy, and operational CO₂e. It supports individual estimates, three-scenario comparisons, and CSV batch analysis for up to 500 models.
01Enter inputsProvide model size and training tokens.
02Set assumptionsChoose efficiency, PUE, and grid intensity.
03CompareReview baseline and sensitivity scenarios.
04ExportDownload the analysis results.
Technical work
Estimation, batch analysis, visualization, and reporting.
Estimation pipeline that separates IT energy from total facility energy
Validated batch analysis for up to 500 models
Three-scenario sensitivity comparisons and interactive Plotly visualizations
CSV, JSON, and PDF exports supported by 14 automated tests and CI