Sentence Transformers

Earlier we demonstrated how to generate embeddings and then search them using OpenAI, but this can also be done using the Sentence Transformers available on Hugging Face.

Setup

Start by connecting to your instance:

psql 'postgresql://postgres:<your-password>@<your-host>:5432/postgres'

vectorize.table() works with ANY model that can be loaded via the SentenceTransformers() API so long as it does not require any additional code execution (which includes most open source sentence transformers).

To get started, setup a products table. Copy from the example data provided by the extension.

CREATE TABLE products (LIKE vectorize.example_products INCLUDING ALL);
INSERT INTO products SELECT * FROM vectorize.example_products;

SELECT * FROM products LIMIT 2;

 product_id | product_name |                      description                       |        last_updated_at
------------+--------------+--------------------------------------------------------+-------------------------------
          1 | Pencil       | Utensil used for writing and often works best on paper | 2023-07-26 17:20:43.639351-05
          2 | Laptop Stand | Elevated platform for laptops, enhancing ergonomics    | 2023-07-26 17:20:43.639351-05

Initializing a job

Create a job to vectorize the products table. We’ll specify the tables primary key (product_id) and the columns that we want to search (product_name and description).

SELECT vectorize.table(
    job_name    => 'product_search_hf',
    "table"     => 'products',
    primary_key => 'product_id',
    columns     => ARRAY['product_name', 'description'],
    schedule    => 'realtime',
    transformer => 'sentence-transformers/all-MiniLM-L6-v2'
);

Private models from Hugging Face

If you’ve uploaded a private model to Hugging Face, you can still host it on Tembo Cloud. Simply reference your Hugging Face org and model name, and pass the API key in as an arg to vectorize.table().

SELECT vectorize.table(
    job_name    => 'product_search_hf',
    "table"     => 'products',
    primary_key => 'product_id',
    columns     => ARRAY['product_name', 'description'],
    transformer => 'my-hugging-face-org/my-private-model',
    schedule    => 'realtime',
    args        => '{"api_key": "hf_my_private_api_key"}'
);

Searching the embeddings

Then search, again specifying the job_name as define above and the columns you want to return.

SELECT * FROM vectorize.search(
    job_name        => 'product_search_hf',
    query           => 'accessories for mobile devices',
    return_columns  => ARRAY['product_id', 'product_name'],
    num_results     => 3
);

                                       search_results
---------------------------------------------------------------------------------------------
 {"product_id": 13, "product_name": "Phone Charger", "similarity_score": 0.8147814132322894}
 {"product_id": 6, "product_name": "Backpack", "similarity_score": 0.7743061352550308}
 {"product_id": 11, "product_name": "Stylus Pen", "similarity_score": 0.7709902653575383}

Directly generating embeddings

You can also generate embeddings directly using vectorize.encode().

select vectorize.encode(
    input => 'mobile electronic accessories',
    model => 'sentence-transformers/all-MiniLM-L6-v2'
);

This can be used in a more complex query, such as:

SELECT
    product_name,
    description,
    1 - (
        my_search_project_embeddings <=>
        vectorize.encode('mobile electronic devices', 'sentence-transformers/all-MiniLM-L6-v2')::vector
    ) as similarity
FROM products
ORDER by similarity DESC
LIMIT 3;