Quick and Dirty Vector DB

BabyVec

'BabyVec' is intended to provide a simple API around common operations over embeddings of text.

It supports an optional HTTP interface via uvicorn and FastAPI.

Some examples of usage:

Building an embedding 'store' for later use in an index.

Code: ./examples/embed_python_documentation.py

Since it is useful to compute embeddings on a beefy machine for later use on a smaller one, the computation of embeddings is generally treated separately from building an index from those embeddings.

Building a searchable embedding 'index'.

Code: ./examples/search_python_documentation.py

Assuming the Python documentation embeddings were created via example 1, this loads them into a searchable index. This script runs a 'prompt / response' flow where the user's input is searched against the Python documentation.

Next steps

testing at larger volumes of data, particularly when all computed embeddings will not fit in RAM
testing on GPU

{
  "by": "ltengelis",
  "descendants": 1,
  "id": 40249564,
  "kids": [
    40249565
  ],
  "score": 1,
  "time": 1714754256,
  "title": "Quick and Dirty Vector DB",
  "type": "story",
  "url": "https://github.com/lummm/babyvec"
}

{
  "author": "Black-Tusk-Data",
  "date": null,
  "description": "Tiny vector embedding interface for natural language processing - Black-Tusk-Data/babyvec",
  "image": "https://opengraph.githubassets.com/33d9b7e7df4a07e16ab3fb755ac9699449031eb1e3a337ba989298a39759ab46/Black-Tusk-Data/babyvec",
  "logo": "https://logo.clearbit.com/github.com",
  "publisher": "GitHub",
  "title": "GitHub - Black-Tusk-Data/babyvec: Tiny vector embedding interface for natural language processing",
  "url": "https://github.com/Black-Tusk-Data/babyvec"
}

{
  "url": "https://github.com/Black-Tusk-Data/babyvec",
  "title": "GitHub - Black-Tusk-Data/babyvec: Tiny vector embedding interface for natural language processing",
  "description": "BabyVec 'BabyVec' is intended to provide a simple API around common operations over embeddings of text. It supports an optional HTTP interface via uvicorn and FastAPI. Some examples of usage:  Building an...",
  "links": [
    "https://github.com/Black-Tusk-Data/babyvec",
    "https://github.com/lummm/babyvec"
  ],
  "image": "https://opengraph.githubassets.com/33d9b7e7df4a07e16ab3fb755ac9699449031eb1e3a337ba989298a39759ab46/Black-Tusk-Data/babyvec",
  "content": "<div><article><p></p><h2>BabyVec</h2><a target=\"_blank\" href=\"https://github.com/Black-Tusk-Data/babyvec#babyvec\"></a><p></p>\n<p>'BabyVec' is intended to provide a simple API around common operations over embeddings of text.</p>\n<p>It supports an optional HTTP interface via uvicorn and FastAPI.</p>\n<p>Some <a target=\"_blank\" href=\"https://github.com/Black-Tusk-Data/babyvec/blob/main/examples\">examples</a> of usage:</p>\n<ol>\n<li>Building an embedding 'store' for later use in an index.</li>\n</ol>\n<p>Code: <a target=\"_blank\" href=\"https://github.com/Black-Tusk-Data/babyvec/blob/main/examples/embed_python_documentation.py\">./examples/embed_python_documentation.py</a></p>\n<p>Since it is useful to compute embeddings on a beefy machine for later use on a smaller one, the computation of embeddings is generally treated separately from building an index from those embeddings.</p>\n<ol>\n<li>Building a searchable embedding 'index'.</li>\n</ol>\n<p>Code: <a target=\"_blank\" href=\"https://github.com/Black-Tusk-Data/babyvec/blob/main/examples/search_python_documentation.py\">./examples/search_python_documentation.py</a></p>\n<p>Assuming the Python documentation embeddings were created via example 1, this loads them into a searchable index.  This script runs a 'prompt / response' flow where the user's input is searched against the Python documentation.</p>\n<p></p><h2>Next steps</h2><a target=\"_blank\" href=\"https://github.com/Black-Tusk-Data/babyvec#next-steps\"></a><p></p>\n<ul>\n<li>testing at larger volumes of data, particularly when all computed embeddings will not fit in RAM</li>\n<li>testing on GPU</li>\n</ul>\n</article></div>",
  "author": "",
  "favicon": "https://github.githubassets.com/favicons/favicon.svg",
  "source": "github.com",
  "published": "",
  "ttr": 28,
  "type": "object"
}