Hi
It takes a bit to load the model, but then works at around 8000-12000 characters/10ms, which is 2-3x faster than the required speed. Have you made repeated calls to language detection within a single launch of a program? Because each launch of a program loads the model again.
There was no mention of time limits to load the model and the only time requirement was about the speed of processing text. Furthermore, there is specifically a "resources" folder in the required project structure (which implies that loading might be necessary). So I (along with many other participants) expected the loading time to not count.
I'd kindly ask you to re-evaluate my (and others') solution, especially that the results were delayed specifically to let participants to respond to issues, but I never received any issue until this one of yours.