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:

	FastEmbed	sentence-transformers
Install size	~100MB	~2GB+
Dependencies	ONNX Runtime	PyTorch
Performance	Optimized inference	Standard
Quality	Similar	Similar

Configuration

%% Using virtualenv (recommended)
{ok, State} = barrel_embed:init(#{
    embedder => {fastembed, #{
        venv => "/absolute/path/to/.venv",
        model => "BAAI/bge-small-en-v1.5",     % default
        timeout => 120000                       % default, ms
    }}
}).

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

Options

Option	Type	Default	Description
venv	string	undefined	Path to virtualenv (recommended)
python	string	"python3"	Python executable (if no venv)
model	string	"BAAI/bge-small-en-v1.5"	Model name
timeout	integer	120000	Timeout in milliseconds

Supported Models

Model	Dimensions	Notes
BAAI/bge-small-en-v1.5	384	Default, fast
BAAI/bge-base-en-v1.5	768	Good balance
BAAI/bge-large-en-v1.5	1024	Highest quality
sentence-transformers/all-MiniLM-L6-v2	384	Popular lightweight
nomic-ai/nomic-embed-text-v1.5	768	Good general-purpose

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