A helper called a tokenizer breaks a sentence into small pieces. Each piece gets an ID number, like a library catalogue number that helps the computer find the right entry.
The website sends your message to OmegaMind. The model has received the question, but it has not produced an answer yet.
Think of the CPU as a busy kitchen. Each of the 16 bars is one round of thinking (a "layer"). Every round runs the same 5 small jobs. After round 16, one last job scores what could come next.
The grid is a peek at the real maths: the CPU multiplies the Input numbers by learned values and keeps a running total. Two workers share the job, and each handles 8 numbers at once.
RAM is the computer's short-term desk space. The bar shows where the model's numbers sit: the lookup table (left), the 16 rounds (middle) and the final scoring table (right).
Saved information reused between steps · simplified viewEach column is one token. The model keeps notes about earlier tokens (the "KV cache") so it never repeats old work.
saved earliernewly savedbeing read now
OmegaMind compares several possible pieces of text, then chooses one. It repeats this process to build a complete answer.