01 · Generation

Token generation loop

Follow a response one token and one model pass at a time.

Open demo →02 · Generation

Temperature

Reshape a probability distribution from focused to exploratory.

Open demo →03 · Representation

Embedding space

Explore semantic neighborhoods and vector arithmetic.

Open demo →04 · Architecture

Attention map

Compare how different heads connect tokens.

Open demo →05 · Input

Tokenizer playground

Split text into model-readable pieces, IDs, and bytes.

Open demo →06 · Generation

Top-k & top-p

Watch candidate filtering change the next-token pool.

Open demo →07 · Inference

KV cache

See cached attention state reduce repeated computation.

Open demo →08 · Architecture

Transformer block

Step a token through attention, normalization, and the MLP.

Open demo →09 · Systems

RAG pipeline

Retrieve evidence, build context, and ground an answer.

Open demo →10 · Architecture

Expert routing

Route each token through a sparse mixture of experts.

Open demo →11 · Optimization

Quantization

Trade numerical precision for memory and speed.

Open demo →12 · Training

Fine-tuning

Watch loss fall as task behavior changes over epochs.

Open demo →