search

karpathy / llama2.c

Inference Llama 2 in one file of pure C
MIT License
17.18k stars 2.04k forks source link

Mojo version? #396

Open MackNcD opened 1 year ago

MackNcD commented 1 year ago

https://github.com/tairov/llama2.mojo

Imagine 250x speed on the original...

espio999 commented 1 year ago

Imagine 250x speed on the original...

Probably you've noticed, this original means llama2.py, not llama2.c, I am interested in Mojo, though it requires minimum 8GiB RAM only for SDK.

kroggen commented 1 year ago

The C version is faster when using multi-threading :fire:

performance comparison