Hello,

Recently I have been trying to incorporate on-device AI Text Embeddings into my Flutter app. Since there is no straight-forward library in Flutter I wanted to use flutter_rust_bridge. The respective crate can be found under: crates.io: Rust Package Registry.

My idea was to use the tokenizers crate for the tokenization process, implement it in rust, generate the respective dart bindings using flutter_rust_bridge_codegen generate and later run an embedding model directly in dart.

The problem is that the flutter_rust_bridge crate has limited support for automatic third-party translation when generating the bindings. I tiredly tried the manual approach but I don't manage to make it work - I am quite overwhelmed since it is new for me. The bindings are always faulty, though I know that it should be possible:

If it errors, then yes the experimental automatic translation cannot understand some part of it yet. But given that the API amount may not be large, the manual approach may not take a large amount of time (via the library OP)

My API is fairly simple: Load a tokenizer from a given JSON file and expose one function tokenize that calls: tokenizer.encode("Hey there!", false)?;

Greetings and many thanks!

1 post - 1 participant

Read full topic