GitHub actions

[martindevans]

Added some simple CI which builds the project and runs tests in debug and release mode on Linux, Windows and MacOS.

I've added an action to download a model into the test project as part of the build. This file is not comitted to the repo, but it's cached in the CI so it should be fast after the first time.

This also incorporates a fix to the netstandard2.0 build (formerly #73) so that the CI doesn't immediately fail!

[martindevans]

I've added a single test which simply loads and disposes a model (previously there were no tests). I'm not sure why this is failing on Linux and MacOS, probably an issue with the way the test project was set up but I'm not sure what.

[martindevans]

Someone on Discord (jlsantiago) did some investigation into why this is failing on Mac. No conclusion yet.

Here's the conversation for reference: https://discord.com/channels/1106946823282761851/1106947264938790972/1137050602711560252

[saddam213]

It looks like there was some issues Marshaling the tensor_split affecting Linux and Mac

I have commited a fix 2b5efa0 that should work on all platforms, can confirm Winows and Linux, but I dont have a recent compilation of the Mac lib to test

However 99% sure this will also work for mac also

EDIT: Actions now successful for Linux
https://github.com/saddam213/LLamaSharp/actions/runs/5769657251

[martindevans]

@saddam213 thanks for investigating this 👍

Would you mind opening a PR into my fork for this fix?

[saddam213]

I think this branch predates the tensor_split change in llama.cpp

[martindevans]

I think this branch predates the tensor_split change in llama.cpp

@saddam213 I've rebased this branch onto master, so it should be fully up to date now.

Note: this has broken the CI on Windows, since the DLLs are out of date right now

[martindevans]

I'm going to merge this PR and the MacOS binaries update by @SignalRT (#77). After that we'll have CI running and we can see what the status is with cross platform support and binary versions etc.