Need a framework tutorial for newcomer

[Yangxiaoz]

Nowadays, llama.cpp is a truly awesome project, but as it supports more and more features, some concepts might be quite confusing for newcomers who haven't been exposed to related fields before.

So I shifted my focus to learning the fundamental concepts related to GGML. However, since I initially lacked a clear and accurate understanding of concepts like “context, tensor, and backend” I ended up spending a lot of time trying to grasp them.

After gaining an understanding of the concepts by reading the source code, I realized that GGML and llama.cpp are truly meaningful and impressive projects. So, would it be possible to add some basic framework tutorials to help more motivated newcomers participate more easily?

I think that doing so could enable llama.cpp and GGML reach a broader stage.

[FSSRepo]

It would be great, this library has a lot of potential. The simple example is a good starting point to learn the basics of GGML and how to perform a simple matrix multiplication.

[cmdr2]

@Yangxiaoz I agree that ggml would benefit greatly from more official tutorials and API documentation. I found this intro on huggingface helpful (but some of the code has changed slightly since then).

If I may plug my posts, I've been writing a series of beginner-friendly tutorials on ggml. More posts coming soon.

Part 1 - covers a simple ggml program to add tensors (on the CPU and CUDA) - https://cmdr2.org/notes/2025/02/1739822447/

Part 2 - writes a simple neural network using ggml - https://cmdr2.org/notes/2025/02/1739897089/

I'm a beginner to ggml as well, and these posts are my way of consolidating my understanding (as I learn about them). I'm working on Part 3, which will talk about reducing memory usage during inference (i.e. quantization and float16).