0 gpu usage on textcat-multilabel

I have classified 1k examples with:

prodigy textcat.teach textcat_spam blank:en ./input.jsonl --label SPAM --patterns ./spam-patterns.jsonl

and then tried to run:

prodigy train spam-classifier-v2 --textcat-multilabel textcat_spam --eval-split 0.2 --lang en --gpu-id 1

which complained about cupy not being installed so I installed it (cuda 12.3 on host, so cuda12x package)

it still is only training on one CPU core, why? thanks

Welcome to the forum @punctal !

Could you please run your train command with PRODIGY_LOGGING=basic and share the stdout please?
The commands look correct to me.