Practical methodology for understanding language model architectures

A guide on how to study and document architectures of modern language models. The process starts with official technical reports but leverages the availability of open weights on platforms like Hugging Face and the Python transformers library to directly inspect configuration files and reference implementations. The approach is primarily manual, enabling researchers to work effectively with open-source models, while proprietary systems like ChatGPT, Claude, and Gemini remain inaccessible. The author emphasizes that studying these architectures manually remains a valuable exercise for understanding how these systems work internally.