From strings to tokens to embeddings

Fill in tokens and token IDs, assign each token ID a 2D embedding, and watch the vocabulary points move in the plane. In the LLM sense, token IDs index rows of the embedding matrix, and each token maps to one such row.

Step 1 enter a token

Step 2 map tokens to token IDs

Step 3 drag points to reshape the embedding space

Try a Sentence

Type a short sentence and the page will map each word to its token and token ID.

Token Sequence

Token IDs

Token-ID Embeddings

Each row contains the token string, its token ID, and the matching 2D embedding vector.

Embedding Plane

Word labels are drawn directly on the points.

Drag any point to update the embedding matrix. Click a matrix row or a point to focus a token ID.

Token-ID point

Focused token ID

Shared token ID