pgvector support for Elixir
Add this line to your application’s mix.exs under deps:
{:pgvector, "~> 0.3.0"}And follow the instructions for your database library:
Or check out some examples:
- Embeddings with OpenAI
- Binary embeddings with Cohere
- Sentence embeddings with Bumblebee
- Hybrid search with Bumblebee (Reciprocal Rank Fusion)
- Sparse search with Bumblebee
- Horizontal scaling with Citus
- Bulk loading with
COPY
Create lib/postgrex_types.ex with:
Postgrex.Types.define(MyApp.PostgrexTypes, Pgvector.extensions() ++ Ecto.Adapters.Postgres.extensions(), [])And add to config/config.exs:
config :my_app, MyApp.Repo, types: MyApp.PostgrexTypesCreate a migration
mix ecto.gen.migration create_vector_extensionwith:
defmodule MyApp.Repo.Migrations.CreateVectorExtension do use Ecto.Migration def up do execute "CREATE EXTENSION IF NOT EXISTS vector" end def down do execute "DROP EXTENSION vector" end endRun the migration
mix ecto.migrateYou can now use the vector type in future migrations
create table(:items) do add :embedding, :vector, size: 3 endAlso supports :halfvec, :bit, and :sparsevec
Update the model
schema "items" do field :embedding, Pgvector.Ecto.Vector endAlso supports Pgvector.Ecto.HalfVector, Pgvector.Ecto.Bit, and Pgvector.Ecto.SparseVector
Insert a vector
alias MyApp.{Repo, Item} Repo.insert(%Item{embedding: [1, 2, 3]})Get the nearest neighbors
import Ecto.Query import Pgvector.Ecto.Query Repo.all(from i in Item, order_by: l2_distance(i.embedding, ^Pgvector.new([1, 2, 3])), limit: 5)Also supports max_inner_product, cosine_distance, l1_distance, hamming_distance, and jaccard_distance
Convert a vector to a list or Nx tensor
item.embedding |> Pgvector.to_list() item.embedding |> Pgvector.to_tensor()Add an approximate index in a migration
create index("items", ["embedding vector_l2_ops"], using: :hnsw) # or create index("items", ["embedding vector_l2_ops"], using: :ivfflat, options: "lists = 100")Use vector_ip_ops for inner product and vector_cosine_ops for cosine distance
Register the extension
Postgrex.Types.define(MyApp.PostgrexTypes, Pgvector.extensions(), [])And pass it to start_link
{:ok, pid} = Postgrex.start_link(types: MyApp.PostgrexTypes)Enable the extension
Postgrex.query!(pid, "CREATE EXTENSION IF NOT EXISTS vector", [])Create a table
Postgrex.query!(pid, "CREATE TABLE items (embedding vector(3))", [])Insert a vector
Postgrex.query!(pid, "INSERT INTO items (embedding) VALUES ($1)", [[1, 2, 3]])Get the nearest neighbors
Postgrex.query!(pid, "SELECT * FROM items ORDER BY embedding <-> $1 LIMIT 5", [[1, 2, 3]])Convert a vector to a list or Nx tensor
vector |> Pgvector.to_list() vector |> Pgvector.to_tensor()Add an approximate index
Postgrex.query!(pid, "CREATE INDEX ON items USING hnsw (embedding vector_l2_ops)", []) # or Postgrex.query!(pid, "CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100)", [])Use vector_ip_ops for inner product and vector_cosine_ops for cosine distance
Create a vector from a list
vec = Pgvector.new([1, 2, 3])Or an Nx tensor
vec = Nx.tensor([1.0, 2.0, 3.0]) |> Pgvector.new()Get a list
list = vec |> Pgvector.to_list()Get an Nx tensor
tensor = vec |> Pgvector.to_tensor()Create a half vector from a list
vec = Pgvector.HalfVector.new([1, 2, 3])Or an Nx tensor
vec = Nx.tensor([1.0, 2.0, 3.0], type: :f16) |> Pgvector.HalfVector.new()Get a list
list = vec |> Pgvector.to_list()Get an Nx tensor
tensor = vec |> Pgvector.to_tensor()Create a sparse vector from a list
vec = Pgvector.SparseVector.new([1, 2, 3])Or an Nx tensor
vec = Nx.tensor([1.0, 2.0, 3.0]) |> Pgvector.SparseVector.new()Or a map of non-zero elements
elements = %{0 => 1.0, 2 => 2.0, 4 => 3.0} vec = Pgvector.SparseVector.new(elements, 6)Note: Indices start at 0
Get the number of dimensions
dim = vec |> Pgvector.SparseVector.dimensions()Get the indices of non-zero elements
indices = vec |> Pgvector.SparseVector.indices()Get the values of non-zero elements
values = vec |> Pgvector.SparseVector.values()Get a list
list = vec |> Pgvector.to_list()Get an Nx tensor
tensor = vec |> Pgvector.to_tensor()Lists must be converted to Pgvector structs for Ecto distance functions.
# before l2_distance(i.embedding, [1, 2, 3]) # after l2_distance(i.embedding, ^Pgvector.new([1, 2, 3]))View the changelog
Everyone is encouraged to help improve this project. Here are a few ways you can help:
- Report bugs
- Fix bugs and submit pull requests
- Write, clarify, or fix documentation
- Suggest or add new features
To get started with development:
git clone https://github.com/pgvector/pgvector-elixir.git cd pgvector-elixir mix deps.get createdb pgvector_elixir_test mix testTo run an example:
cd examples/loading mix deps.get createdb pgvector_example mix run example.exs