Make msvc compile and improve CI

[dagelf]

Fixes #200 breaking msvc.

[dagelf]

😅 Huggingface is straining.

OSError: We couldn't connect to 'https://huggingface.co/' to load this file, couldn't find it in the cached files and it looks like gpt2 is not the path to a directory containing a file named config.json.

I promise it works! And removes the warnings for clang too

~/llm.c$ clang -Ofast -Wno-unused-result -march=native -fopenmp -DOMP   train_gpt2.c -lm -lgomp -o train_gpt2
~/llm.c$ cc -Ofast -Wno-unused-result -march=native -fopenmp -DOMP   train_gpt2.c -lm -lgomp -o train_gpt2
~/llm.c$ 

[dagelf]

Also resolves #217 ... partially

[karpathy]

the shakespeare download is super tiny, is it causing issues in the CI? should we be paying the cost of merging that to the repo for it?
compared to that, the huggingface download is a lot larger.

[dagelf]

Chances are it was the GPT2 download, maybe I didn't stare at the log hard enough... but:

https://status.huggingface.co/

[dagelf]

Of course you're right. This is why #215 failed:

[dagelf]

Still, so far it's one less download costing just 1M ... and it fixes msvc compile.

Should I move all the output .bin files to data/ too? Currently they clutter the root.

[azret]

add

/.vs

as well please for visual studio

[karpathy]

I don't want to clutter the repo with shakespeare if we can avoid it. It's a pet peeve to git clone a repo and have to wait long seconds/minutes for the download.

[karpathy]

Let's not move the bin files right now, let's try to keep this PR focused and minimal

[dagelf]

Impressive that this repo is still under 2M, including all history. New PR incoming.