Chat Bot implemented in pure Rust using Mistral-7B with HuggingFace/Candle over Axum Websockets and a Leptos (Wasm) frontend!
If you are using Debian/Ubuntu you should be able to get up and running with a single make init
.
make prod
runs both Frontend (Leptos) and Backend (Axum) in with the --release
flag.
View make help
to see all commands.
Websocket defaults to 127.0.0.1:3000
unless /backend/.env
file includes:
IPV4=
# and/or
PORT=
Default model options can be configured with /backend/config_model.yaml
.
In order to configure running the model with Cuda
adjust
/backend/config_model.yaml
to say cpu: false
.
And, enable the cuda feature flags:
cargo add candle-core -F "cuda"
cargo add candle-transformers -F "cuda"
Default inference options can be configured with /backend/config_inference.yaml
Frontend defaults to localhost:8080