Export llama without llama

[python273]

Now only need torch + numpy, GPU not needed, no need to clone https://github.com/facebookresearch/llama

7B works, 13B doesn't work for some reason. But, I have llama2c format loader for Tinygrad and it works there, so there must be some bug in run.c?

[sidsarasvati]

You wanna update the README as well? I can test it out on my M1 mac

[python273]

😅 Seems like there are already #71 and #78. My version works without llama repo, CPU only, supports models bigger than 7B

[karpathy]

Thank you I will take it, and change some of the docs too. Let's worry about the 13B later. Strange that it would work in tinygrad but not here.