RAGEngineering Hub
LAB · embedding-labRuns in your browser

Embedding Lab

Type two sentences and see how close their vectors are, with the same model the course uses. Then see one golden question scored against every chunk. The model runs in your browser.

Used on RAG-19 and RAG-20

Compare two sentences

  1. Pick a pair and press Compare. Results stay here so you can compare pairs with each other.

bge-small-en-v1.5, q8, passages without a query prefix, as on RAG-20. Compare scores with each other, not with a fixed line.

What to notice

Negation scores about 0.94: opposite advice, almost the same vector. The paraphrase shares no words and still scores about 0.80.

Surprise

In the second panel even the last chunk scores near 0.4. A score only means something next to the others.

Try next

Write a pair from your own work: two product codes, two prices, a sentence and its translation.