How to run a Prithvi-EO fine-tune for crop classification

How it is run today: Self-host open weights or process code

Prithvi-EO is pretrained on HLS and is fine-tunable for crop classification, flood and burn scars, and similar heads. Running that fine-tune is a self-host job: pick a checkpoint, bring labels, train, then serve.

The customer supplies HLS-style 6-band stacks (Blue, Green, Red, Narrow NIR, SWIR1, SWIR2) as GeoTIFF sequences, plus labels if fine-tuning. GPU class is unknown until the variant is chosen.

There is no public hosted fine-tune API for this. IBM enterprise Granite geospatial variants may exist under terms we have not seen.

We can stand up a selected checkpoint and a customer head after a named job. We do not host GPUs today. We do not train on customer data unless that is contracted separately. We do not guarantee classification accuracy.

Upstream

Also in the catalog : Prithvi-EO (IBM–NASA).

Reach out

If this is the job you need, say so. We will not claim we already host it. We will say whether we pass through, stand it up, or cannot do it honestly.