Finally start work on `prodigy-embetter`
koaning opened this issue · comments
vincent d warmerdam commented
python -m prodigy textcat.emb.manual <dataset> <examples.jsonl> --labels --loader --anchors --exclusive
python -m prodigy image.clip.by_text <dataset> <examples.jsonl> --labels --loader --anchors --exclusive --remove-base64
python -m prodigy image.clip.by_image <dataset> <examples.jsonl> --labels --loader --anchors --exclusive --remove-base64
vincent d warmerdam commented
After working on the "frontpage" project, I think this is no longer the best way to go about this. Calculating the embeddings on the fly is expensive and it may be better to have a simple ANN index instead.