Andrej
You're not wrong and it is very tempting but also I've been down this path before. Next thing you know I'll be using argparse, then I'll switch to yaml files,...
addressed in https://github.com/karpathy/nanoGPT/commit/177d5f7dc5f44d6f373cd7767c2a9259d740436e thanks @vgoklani , closing shortly.
good reading thank you for the pointer. Like the compactness of model.py and nice to discover `tyro`
Hi ty - I like the encodings change - I don't like boilerplate gitignores, I like when they're specific to repo - pip installing pytorch would not grab the nightly...
I'd like to have better requirements flow for the repo but don't love this specific approach. Will think on. Ty.
Thank you for the entity None tip. I do want to do this but I think I have to first figure out what config solution I'm happy with. I think...
This is now supported and implemented by guessing the args using something similar: ```python config_keys = [k for k,v in globals().items() if not k.startswith('_') and isinstance(v, (int, float, bool, str))]...
Actually I think this issue is great to keep open, in case anyone investigates nanoGPT in `mps` context. I haven't tried yet.
@itakafu thank you for reporting, i'll add mentions of `mps` to the readme&code.
😂💀 wow, not too bad looking at all, let me play with for a bit...