FastEmbed Provider

Lightweight local embedding using FastEmbed (ONNX-based).

Requirements

# Using virtualenv with uv (recommended)
./scripts/setup_venv.sh
uv pip install fastembed --python .venv/bin/python

# Or install manually
pip install fastembed

Why FastEmbed?

FastEmbed is a lighter alternative to sentence-transformers:

Configuration

{ok, State} = barrel_embed:init(#{
    embedder => {fastembed, #{
        python => "python3",                    % default, fallback only
        model => "BAAI/bge-small-en-v1.5",     % default
        timeout => 120000                       % default, ms
    }}
}).

Options

barrel_embed manages its own Python virtualenv automatically and installs fastembed into it on first use. See Python Virtualenv Setup for the managed venv API.

Supported Models

Example

%% Initialize
{ok, State} = barrel_embed:init(#{
    embedder => {fastembed, #{
        model => "BAAI/bge-small-en-v1.5"
    }}
}).

%% Generate embedding
{ok, Vec} = barrel_embed:embed(<<"Fast and lightweight">>, State).
384 = length(Vec).

%% Batch embedding
{ok, Vecs} = barrel_embed:embed_batch([
    <<"Document 1">>,
    <<"Document 2">>
], State).

When to Use FastEmbed

Choose FastEmbed when:

• Disk space is limited
• PyTorch is not desired
• You need lightweight local inference
• Quality requirements are standard (not maximum)

Choose sentence-transformers (local provider) when:

• You need access to all HuggingFace models
• You're already using PyTorch
• You need maximum compatibility