Chabbo replying to the prompt "Why", trained on The Picture of Dorian Gray
A chatbot for people who are tired of sophisticated AI like ChatGPT and would rather talk to an inane, nonsensical bot trained on their own choice of material.
Written to run on Deta Space but fully functional in local
environments. When the DETA_PROJECT_KEY is set, the application will use Deta
as a backend.
cargo install --git http://github.com/wonderfulspam/chabbo or install to your
Deta space.
Install just to run the commands defined in the
justfile. Otherwise, the file serves as living documentation of the things you
can do.
Chabbo can swap out its guts on the fly, mostly owing to the fact that constructing a Markov chain is a couple of dozen orders of magnitude simpler than actually training an ML model.
The default corpus is The Picture of Dorian Gray as found on Project Gutenberg. To use a different corpus, follow the instructions in the web UI.
A corpus should consist of line-delimited sentences without superfluous
punctuation. The scripts/ folder contains scripts for extracting text from
Telegram and Messenger chat exports as well as for tidying up books found on
Project Gutenberg.
The text generation itself is a simple Markov chain. It runs as an Axum app with a minimal amount of vanilla JS to facilitate communication + PicoCSS to make the application look vaguely modern. Communication with Deta is handled via wondefulspam/deta-rust, a fork of jnsougata/deta-rust.
- Be generic over file system & deta backends
- Allow changing/adding corpora
- Allow changing Markov chain order on the fly
- Build on Deta instead of shipping binary directly (awaiting upstream fix)
- Improve error handling to improve resilience
- Replace the trainwreck of a frontend with eg. a Svelte app
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in in this project by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.