You can download this repository, and remove some models ngrams locally (so that the final binary would be lighter), but remember that an ngram model len must be the same for all of the languages of the script (Hans, Hant, Kore, Jpan must be considered as the same script).
git clone https://github.com/RoDmitry/langram_models.git
git clone https://github.com/RoDmitry/langram.git
cd ./langram_models
# for big-endian targets: in Cargo.toml enable `rkyv::big_endian` feature
cargo run --release
Trained on OpenLID (201 languages).