Llama 2 inference in Zig

1. Start and get inside the Docker container:

   cd infra-dev/
   docker-compose up -d
   docker-compose exec -it llama2 bash

2. Download models:

   curl -L https://huggingface.co/karpathy/tinyllamas/resolve/main/stories15M.bin -o ../models/TinyLlama-15M.bin

   See model.py for more details about how the .bin file was exported.
3. Inference:

   zig build run -- ../models/TinyLlama-15M.bin ../llama2.c/tokenizer.bin

   Or:

   zig build-exe ./src/main.zig -O ReleaseFast -lc
   ./main ../models/TinyLlama-15M.bin ../llama2.c/tokenizer.bin

   Output:

   Hello darkness, my old friend, the sun. He is very hot and he needs to cool down. He looks around and sees a big tree. He thinks it looks like a good place to rest.
   He climbs up the tree and looks around. He sees a big, green tree with lots of leaves. He thinks it looks like a good place to rest. He climbs up the tree and sits on a branch. He feels the cool breeze on his face.
   He looks around and sees a little girl. She is playing with her doll. She has long hair and a pink dress. She looks at him and smiles. She says, "Hello, mister. Do you like my tree?"
   The old man nods and says, "Yes, I do. It is very nice. Do you want to play with me?"
   The little girl nods and says, "Yes, I do. I like your tree. It is very big and green. Can I sit with you?"
   The old man says, "Sure, you can sit with me. But be careful, don't touch my tree. It is very old and fragile. It can break easily."
   The little girl says,
   

• llama2.c
• llama2.rs
• llama2.c-for-dummies