AMIP Inference (HPX64)

Prerequisites:

• Install cBottle
• Download the cBottle-3d.zip checkpoint from NGC.

Tutorial

You can run an SST-conditioned (AMIP) inference like this

python scripts/inference_coarse.py \
    --sample.min_samples 1 \
    --dataset amip \
    cBottle-3d.zip \
    inference_output

Note

If not already download, this will first download the SST data from ESGF.

If you plan to run a big parallel job, you can prefetch this first by runing

python3 scripts/download_amip_sst.py

By default this will download to ~/.cache/cbottle, but if desired you can configure the environment variables to point to the downloaded data:

export AMIP_MID_MONTH_SST="/path/to/data"

To persist this setting across multiple sessions, you can put this in your .bashrc, submission scripts, or a dotenv file.