← Back to paper

Generation Visualizer

Watch MOSAIC build molecules token-by-token. Each tokenizer encodes a different level of hierarchical structure — from flat partition blocks to typed compositional trees.

H-SENT Hierarchical SENT

Partition blocks with explicit cross-community edges. Each community's atoms are serialized via a SENT walk with back-edge brackets.

H-SENT autoregressive generation

HDT Hierarchical DFS Tokenization

Full tree structure via depth-first traversal with ENTER/EXIT nesting tokens. Intra- and inter-community edges are captured as back-edges.

HDT autoregressive generation

HDTC HDT with Composition

The most expressive scheme: typed community nodes (ring, functional group, singleton), super-graph of inter-community bonds, and atom-level detail within each block.

HDTC autoregressive generation

H-SENT

H-SENT autoregressive generation

HDT

HDT autoregressive generation

HDTC

HDTC autoregressive generation