Running `quick start on CPU` on Macbook Pro M2 about llm.c

[gordicaleksa]

Hey @full-stack-ai ! There are 2 implementations in llm.c:

1. PyTorch reference implementation which is inside train_gpt2.py -> we only use it as a sanity check that our C implementation is correct.
2. C/CUDA implementation -> that's what you probably care about, to run it after building using make just run ./train_gpt2 in case you care about the CPU implementation.

Let me know if you need any additional help, your feedback is welcome!

Otherwise feel free to close the issue.

[rosslwheeler]

It's saying that you don't have NVCC which is because you're on a Mac. This is okay - I see the cc line so is it building or are there other errors?

[full-stack-ai]

Nothing seems to happen

[rosslwheeler]

Are the executable files there in the directory? Also, which version of gcc are you using?

[full-stack-ai]

Here is the repo files in the same directory:

And gcc version:

[rosslwheeler]

Everything is there. train_gpt2 is built. You are good to go. Feel free to close this issue. Thanks.

[full-stack-ai]

So why the instruction is mentioned to run make train_gpt2? If python train_gpt2.py trains the model, then the documentation should change.