Load a model to start.

SmolLM2 135M is preselected because it is a small download. Larger models (0.5 to 1B parameters) give clearer readouts in the middle layers.

Each column is one token of your prompt. Rows are the model's layers, from the token embeddings at the bottom to the final layer at the top. Each cell shows the token that layer would predict next, shaded by how confident it is.