EmbeddingViz is a browser-based React app for exploring how embedding models place text, files, tokens, and model outputs in vector space. It runs local UI state, model execution, dimensionality reduction, and plotting in the user's browser.

Use it to compare small embedding runs, inspect selected points, switch between PCA, UMAP, and t-SNE-style projections, and render dense point clouds with WebGL.

Install

git clone https://github.com/tsilva/embeddingviz.git
cd embeddingviz
npm install
npm run dev

Open http://127.0.0.1:5173, or the URL printed by Vite if that port is already in use.

Commands

npm run dev      # start the Vite dev server on 127.0.0.1
npm run build    # type-check and build the production bundle
npm run preview  # preview the production build locally

Notes

• Models are loaded in the browser through @huggingface/transformers; remote Hugging Face model files are allowed and browser cache is enabled.
• There is no backend service or persistent app storage. Run history, selected points, and visibility toggles live in React state.
• Model embedding extraction and projection use Web Workers when available, with a main-thread fallback for non-image extraction.
• Token mode can project up to 50,000 tokenizer vocabulary entries.
• The CLIP preset routes text to the text encoder and images to the vision encoder in the shared CLIP embedding space.

Architecture

License

No license file is currently included.