dev-verma/carbon-emissions-model-card

Model Card

Environmental Impact

Carbon emissions for this training run were estimated using the methodology described in Lacoste et al. (2019), Quantifying the Carbon Emissions of Machine Learning, based on the formulas below.

• Hardware Type: NVIDIA L40S (TDP: 350 W)
• Number of GPUs: 7
• GPU Hours: 377.8
• Power Usage Effectiveness (PUE): 1.53
• Total Energy Consumed: 1416.183 kWh
• Geographical Location: europe-north1
• Grid Carbon Intensity: 120 gCO2eq/kWh
• Carbon Emitted: 169.942 kg CO2eq

Calculation

energy_kWh = (TDP_watts * num_gpus * gpu_hours * PUE) / 1000
           = (350 * 7 * 377.8 * 1.53) / 1000
           = 1416.183 kWh

co2_kg = (energy_kWh * region_carbon_intensity_gCO2_per_kWh) / 1000
       = (1416.183 * 120) / 1000
       = 169.942 kg CO2eq

Training Type

fine-tuning