Embeddings Concept
Visualize how LLMs understand words as mathematical vectors.
Embedding Explorer
Provenance
Model hintDemo word vectors: cosine / dim 24 / hash layout
Points28 words
Current focusNone
Click a point to inspect neighbors in the demo embedding space. Distances are local browser-side demo vectors and can guide API-backed embedding checks. Projection, coloring, distance metric and top-k are fully configurable — everything runs in your browser.
Nearest neighbors
Select a pointtop 10
| Word | Similarity | Action |
|---|---|---|
| Select a point to inspect its nearest neighbors. | ||
Legend
- Search match
- Default word
- Selected
- Group highlight
- Group highlight
- Group highlight
- Derived candidate
Similarity matrix (cosine)
Cells show pairwise cosine similarity (first 28 words). Red = high, blue = low.
Select a word below to highlight its row.
Stats (cosine)
Select a point to see per-word stats (mean similarity, nearest, farthest, uniqueness).
Similarity histogram
0% (dissimilar) → 100% (similar)