Neural Lab · Live

Watch how an LLM finds the answer.

Ask anything — even 2 + 2. Real BPE tokens, real next-token logprobs from a hosted model, real streaming response, and the transformer architecture lighting up layer by layer as your question travels through it.

Ask anything · the network will fire end-to-end

⌘/Ctrl + Enter

Ready

Idle — waiting for a question

Type a question and press Run. Real BPE tokens, real next-token logprobs, real streaming answer.

Transformer architecture · 7 layers · live activation

TokenizeEmbedAttentionFFN + RetrieveDeep LayersPredictGenerate

1 · Tokens of your prompt

— · 0 tok

Type a question above to see real BPE tokens.

2 · Top-5 next-token predictions

—

Predictions will appear once you type.

3 · Streamed answer

0 tok

The model's answer will appear here, token by token, as the network fires.

4 · Pipeline events

0 evt

Events will stream in here as the pipeline runs.