pylate-onnx-export

A CLI tool to export HuggingFace ColBERT models to ONNX format for fast Rust inference with Next-Plaid.

Installation

pip install "pylate-onnx-export @ git+https://github.com/lightonai/next-plaid.git#subdirectory=next-plaid-onnx/python"

Quick Start

Export a Model

# Export a ColBERT model to ONNX format
pylate-onnx-export lightonai/GTE-ModernColBERT-v1

# Export with INT8 quantization (recommended for production)
pylate-onnx-export lightonai/GTE-ModernColBERT-v1 --quantize

# Export to a custom directory
pylate-onnx-export lightonai/GTE-ModernColBERT-v1 -o ./my-models --quantize

Push to HuggingFace Hub

# Export and push to HuggingFace Hub
pylate-onnx-export lightonai/GTE-ModernColBERT-v1 --quantize --push-to-hub myorg/my-onnx-model

# Push as a private repository
pylate-onnx-export lightonai/GTE-ModernColBERT-v1 -q --push-to-hub myorg/my-onnx-model --private

Quantize an Existing Model

colbert-quantize ./models/GTE-ModernColBERT-v1

CLI Options

pylate-onnx-export

Usage: pylate-onnx-export [OPTIONS] MODEL_NAME

Arguments:
  MODEL_NAME  HuggingFace model name or local path

Options:
  -o, --output-dir DIR     Output directory (default: ./models)
  -q, --quantize           Create INT8 quantized model
  --push-to-hub REPO_ID    Push to HuggingFace Hub
  --private                Make Hub repository private
  --help                   Show help message

colbert-quantize

Usage: colbert-quantize [OPTIONS] MODEL_DIR

Arguments:
  MODEL_DIR  Directory containing model.onnx

Options:
  --help     Show help message

Output Structure

The tool creates a directory with the following files:

models/<model-name>/
├── model.onnx                        # FP32 ONNX model
├── model_int8.onnx                   # INT8 quantized (with --quantize)
├── tokenizer.json                    # Tokenizer configuration
└── config_sentence_transformers.json # Model metadata (embedding_dim, etc.)

Supported Models

Any PyLate-compatible ColBERT model from HuggingFace can be exported:

Python API

from colbert_export import export_model, quantize_model

# Export a model
output_dir = export_model(
    model_name="lightonai/GTE-ModernColBERT-v1",
    output_dir="./models",
    quantize=True,
)

# Or quantize an existing model
quantize_model("./models/GTE-ModernColBERT-v1")

Usage with Next-Plaid API

After exporting a model, you can use it with the Next-Plaid API:

# Start API with the exported model
docker run -d \
  -p 8080:8080 \
  -v ~/.local/share/next-plaid:/data/indices \
  -v ./models:/models:ro \
  ghcr.io/lightonai/next-plaid-api:model \
  --model /models/GTE-ModernColBERT-v1

Or use a model from HuggingFace Hub (auto-downloaded):

docker run -d \
  -p 8080:8080 \
  -v ~/.local/share/next-plaid:/data/indices \
  -v next-plaid-models:/models \
  ghcr.io/lightonai/next-plaid-api:model \
  --model lightonai/GTE-ModernColBERT-v1-onnx

License

Apache-2.0