One binary for any goruut use

[thewh1teagle]

Go usually compiled into tiny binary we can have everything in single binary instead of compile multiple things and running multiple scripts
Regarding embedding of models i would avoid that anyway and let the user wget what he need

[neurlang]

Every binary which does inference (like the testing binaries: backtest, homotest, phondephontest, plus production binary: goruut) will be fat (150+MB) because it bundles all the models. For these, this solution will make sense.
Every binary which does training-only (analysis2,...) is slim and will be best left as separate binary in cmd to speed up compilation. Plus that code is not needed for production use, so that's another valuable space we save in production.

[neurlang]

Regarding embedding of models we don't have a guaranteed caching directory. Imagine running in browser compiled into wasm (works now) or running in read only filesystem machine where there's really no place to download models to.

[neurlang]

Not to mention offline use.

[thewh1teagle]

Regarding embedding of models we don't have a guaranteed caching directory. Imagine running in browser compiled into wasm (works now) or running in read only filesystem machine where there's really no place to download models to.

Then it can be feature flag and identify what's available in runtime

[thewh1teagle]

Can we add build.sh instead?

[thewh1teagle]

I would keep this issue for future until some solutions will be find
it will be nice to simplify the end user usage as well the cycle of adding language