Skip to content

Latest commit

 

History

History
42 lines (29 loc) · 1.93 KB

README.md

File metadata and controls

42 lines (29 loc) · 1.93 KB
Raw

AI-Carbon-Counter

Count the amount of carbon used for creating an AI model using Impact framework and display in Hugging face

Required input params

  • timestamp - When the training for started
  • duration - how much took to complete the training in seconds
  • cloud/instance-type - Cloud instance name understandable by cloud-metadata plugin
  • cloud/vendor - Cloud vendor like azure, gcp, aws understandable by cloud-metadata plugin
  • cloud/region - Cloud regin where the machines were located cloud/region understandable by cloud-metadata plugin
  • cpu/utilization - If you know approximate usage percentage, otherwise go with 100

Other information

WattTime API requires activation of subscription before usage. Please refer to the WattTime website for more information.

Create a .env file in the IF project root directory. This is where you can store your WattTime authentication details. Your .env file should look as follows:

Required Parameters:

WATT_TIME_USERNAME: <your-username>
WATT_TIME_PASSWORD: <your-password>

How to run

install node

npm install -g "@grnsft/if-plugins"

npm install -g "@grnsft/if-unofficial-plugins"

ie --manifest ./ai_carbon_counter.yml --output output

The final output will be available in output file. Look for carbon variable for the carbon usage in grams.

Limitations

  • duration to be within 32 days (WattTime plugin).
  • Emissions are aggregated for every 5 minutes regardless of the granularity of the inputs (WattTime plugin).
  • Limited instance types support (cloud-metadata plugin)