I’m so excited to show the updated version of my latest open-source project here. The previous version was built for Llama 2 and was now updated to support Llama 3.1 8B-Instruct model.

If you are curious like me about how the LLMs (Large Language Models) and transformers work and have delved into conceptual explanations and schematic drawings in the sources but hunger for deeper understanding, then this project is perfect for you too!

You will not only find the details of the Llama architecture but will find explanations of a wide variety of related concepts in the documentation directory. From reading a Pickle, a PyTorch model, a Tiktoken tokenizer model files at byte-by-byte level, to the internals of BFloat16 data type, implementation from scratch of a Tensor structure and mathematical operations including linear algebraic computations.

This project was initially started to learn what an LLM does behind by running and debugging it and was made for experimental and educational purposes only, not for production use.

The goal is to make an experimental project that can perform inference on the Llama 3.1 8B-Instruct model completely outside of the Python ecosystem (using the Go language). Throughout this journey, the aim is to acquire knowledge and shed light on the abstracted internal layers of this technology.

This journey is an intentional journey of literally reinventing the wheel. While reading my journey in the documentation, you will see the details of how Large Language Models work, through the example of the Llama model.

I will be happy if you check out it and comments are welcome! Code and documentation: https://github.com/adalkiran/llama-nuts-and-bolts

And now, the documentation is available on Github Pages: https://adalkiran.github.io/llama-nuts-and-bolts