Earth-2 Correction-Diffusion Model

A guide for generating AI weather forecast using NVIDIA Earth2Studio CorrDiff Model

Required Installations

Install conda in your terminal and make sure python3 --version > 3.10

pip install earth2studio

Activate earth2studio kernel 'conda activate earth2studio'

Verify installation in your Python3 kernel

import earth2studio
print(earth2studio.__version__)

Install submodules NVIDIA Earth-2 Correction Diffusion in NIM

pip install earth2studio[corrdiff]

To ensure that data loads safely

pip install earth2studio[data]

Launching the NIM

To gain access to the CorrDiff model, we need to use the NIM(NVIDIA Inference Microservice). To launch the NIM, a personal API Key is required for the login process. To obtain the API key, register an account with NVIDIA HERE. It is recommended that you paste your key somewhere you have easy access to.

Note that your key value starts with nvapi- and ends with vEg. Your key is only valid for one year. If you ever lose your key, you should delete the old one and generate a new API Key.

Caution

Your personal key should be kept private at all times.

To properly launch the NIM on your NVIDIA pod, download and edit the corrdiff-nim-deployment.yaml file to your username and change the value to your personal API key.

name: NGC_API_KEY
  value: "YOUR API KEY"

To create the deployment, run the following in your terminal:

kubectl create -f corrdiff-nim-deployment-<YOUR NAME>.yaml -n sdsu-shen-climate-lab

Then run: kubectl get pods -n sdsu-shen-climate-lab | grep corrdiff

Tip

It take about 5-10 mins for Kubernetes cluster to download the images(~26GB), keep checking the real-time status of the container in watch mode by running kubectl get pods -n sdsu-shen-climate-lab -w | grep corrdiff.

Once the status shows running: kubectl logs -f deployment/corrdiff-nim-<YOUR NAME> -n sdsu-shen-climate-lab

Run CorrDiff