I've made a visual walk through the machinery inside a large language model: from raw text, to tokens, to vectors, to attention, to the next token.

If you have any comments/questions/remarks/improvements, let me know!