Why Our 13 MB GST Invoice AI Failed — And How a 107 MB Desktop Engine Hit 95.95% Entity F1
DEV Community
Why Our 13 MB GST Invoice AI Failed — And How a 107 MB Desktop Engine Hit 95.95% Entity F1
We compressed a LayoutLM model to 13 MB for phones and it scored 62.6% Entity F1 on real bills. Moving to a 107 MB desktop engine hit 95.95% — fully offline.
0 comments
No comments yet.