A little help for our friends

[SINAPSA-IC]

The effort of developers here is both commendable and honorable.
I as an individual and as a CEO (for whatever it's worth), pledge my full support for the developers and community here, of GPT4All.
I'm only able to assist with

• suggestions for the UX
• C#
• some Prompt Engineering
• 1-2 GB of storage space on our website; not much, but handy for ZIPs with files for LocalDocs

I, for, one would be pleased if lurkers here would give as detailed a feedback as they can - if they can and want, ofc - on any issue.
Things are moving Very fast in this area, nowadays - Gemini 1.5 aside - so I'm Sure that any helping hand to developers here would be useful in a != null degree.

Thank you, all lurkers :)

/dodging stones

[cebtenzzre]

Sure, we would definitely appreciate the help. Right now the C# bindings on master are out-of-date relative to the backend and the python bindings, and the NuGet package does not seem to be packaging the native libraries such that they will be actually found - so the C# bindings are basically not working for anyone. Would you like to take a look?

Feel free to ping me on the Discord if you have any questions.

[SINAPSA-IC]

Hello.

Thank you!

Yes, but I cannot guarantee that I'll understand it right away. But I think that some comments will clear out some of the confusion :)

Still I'm not on Discord...

[cebtenzzre]

Basically PR #1970 broke all of the bindings except for Python, since it made a breaking change to the backend API and the Python bindings are the highest priority at Nomic. This is what had to change in the Python bindings:

• _pyllmodel.py
• gpt4all.py

The C# bindings live here and the CI for them (which builds the NuGet package) is here.

[SINAPSA-IC]

I'm 4 hours in trying to make sense of the code and "guide" left by whoever's broken this C# section or left it for dead for whatever reason, now the last message, original btw, is that the directory "runtimes" does not appear to contain CMakeLists.txt, which I don't know what it is.
By the looks of it, it's FUBAR, excuse my French.
I'll keep looking into it in the coming days.

[cebtenzzre]

the last message, original btw, is that the directory "runtimes" does not appear to contain CMakeLists.txt

Hm, what OS, and what command are you running? AFAICT the build scripts only reference the runtimes directory as a build dir for cmake, not a source dir.

Also, if you'd prefer not to make a Discord account (we have a #gpt4all-bindings channel), I'm also on IRC (cebtenzzre on libera.chat). I just think it would be easier for me to guide you if we had a realtime communication channel.

[SINAPSA-IC]

Hello.

This is a Windows 10, no GPU.

Also, looks like whoever wrote the readme.md instructions and the script benefited from the Windows /10 SDK and cmake being installed on their computer; hasn't mentioned this, so the hint below was not enough to make it all work:

Modify cd gpt4all/gpt4all-backends/CMakeLists.txt adding CXX_STANDARD_20 to llmodel properties.

set_target_properties(llmodel PROPERTIES
                             VERSION ${PROJECT_VERSION}
                             CXX_STANDARD 20 # <---- ADD THIS -----------------------
                             SOVERSION ${PROJECT_VERSION_MAJOR})

The "kompute" message in the image below got me frightened, not one PC here with a GPU; I guess that this one (https://github.com/KomputeProject/kompute/) is needed, too; another implicit assumption, maybe, that kompute does indeed exist on any machine where this script has to run, which is so dead wrong. If I'm successful, which I doubt, I'll write a tutorial of sorts from A to Z, with screen captures and all.

So... nothing really to tell at this point, wish I had something.
After ~10 hours (trial-and -error, StackExchange, MS docs etc.) still at stage 1 of n - getting build_win-msvc to run wholly. A shame, really, but I keep on trying, color me ambitious.

[SINAPSA-IC]

In llama.cpp:

if (APPLE)
    set(LLAMA_KOMPUTE_DEFAULT OFF)
else()
    set(LLAMA_KOMPUTE_DEFAULT ON)
endif()

which triggers the "kompute not found", and cannot advance past this point; changing it to ON needs a non-Windows system which I do not have around.

I've tried putting together a DLL (llibmodel) like Frankenstein's monster, but could not get this method in NativeMethods.cs to start the example code in Readme.md, by setting an EntryPoint:

[DllImport("libllmodel", CallingConvention = CallingConvention.Cdecl, ExactSpelling = true, BestFitMapping = false, ThrowOnUnmappableChar = true)]
[return: NativeTypeName("llmodel_model")]
public static extern IntPtr llmodel_model_create2(
    [NativeTypeName("const char *")][MarshalAs(UnmanagedType.LPUTF8Str)] string model_path,
    [NativeTypeName("const char *")][MarshalAs(UnmanagedType.LPUTF8Str)] string build_variant,
    out IntPtr error);

I really want to help, never give up, never surrender, but as things stand now, stunted and ever back to square 1, I myself could use some.