Implementation:AnswerDotAI RAGatouille Export To Huggingface Hub

Knowledge Sources	AnswerDotAI_RAGatouille Hugging Face Hub Docs
Domains	Model_Export, Model_Distribution, NLP
Last Updated	2026-02-12 12:00 GMT

Overview

Concrete tool for uploading trained ColBERT model checkpoints to the Hugging Face Hub provided by the RAGatouille library.

Description

The export_to_huggingface_hub function packages a local ColBERT model checkpoint and uploads it to a Hugging Face Hub repository. It validates the checkpoint contains a valid ColBERTConfig, optionally re-saves the model to a temporary directory for clean export, and can also produce a Vespa-compatible ONNX file alongside the upload. The function handles authentication errors and repository creation automatically via the HfApi client.

Usage

Use this function after training or fine-tuning a ColBERT model when you want to distribute the model publicly or privately via the Hugging Face Hub. It is typically the final step in a training workflow, enabling other users to load the model with RAGPretrainedModel.from_pretrained().

Code Reference

Source Location

• Repository: AnswerDotAI_RAGatouille
• File: ragatouille/models/utils.py
• Lines: 25-97

Signature

def export_to_huggingface_hub(
    colbert_path: Union[str, Path],
    huggingface_repo_name: str,
    export_vespa_onnx: bool = False,
    use_tmp_dir: bool = False,
) -> None:
    """
    Upload a ColBERT model checkpoint to the Hugging Face Hub.

    Args:
        colbert_path: Local path to the ColBERT model checkpoint directory.
        huggingface_repo_name: Target repository name on HF Hub (e.g. 'username/model-name').
        export_vespa_onnx: If True, also export and upload a Vespa ONNX model.
        use_tmp_dir: If True, re-save model to a temp directory before uploading.
    """

Import

from ragatouille.models.utils import export_to_huggingface_hub

I/O Contract

Inputs

Name	Type	Required	Description
colbert_path	Union[str, Path]	Yes	Local filesystem path to the ColBERT checkpoint directory
huggingface_repo_name	str	Yes	Target Hugging Face Hub repository ID (format: 'username/repo-name')
export_vespa_onnx	bool	No	Whether to also export a Vespa ONNX model alongside the checkpoint (default: False)
use_tmp_dir	bool	No	Whether to re-save the model to a temporary directory before upload (default: False)

Outputs

Name	Type	Description
return	None	No return value; prints status messages to stdout
side effect	HF Hub upload	Model files are uploaded to the specified Hugging Face Hub repository
side effect	Repo creation	Creates the repository on HF Hub if it does not exist

Usage Examples

Basic Upload

from ragatouille.models.utils import export_to_huggingface_hub

# Upload a trained ColBERT model to Hugging Face Hub
export_to_huggingface_hub(
    colbert_path="experiments/colbert/none/2024-01/checkpoints/colbert",
    huggingface_repo_name="myuser/my-colbert-model",
)

Upload with Vespa ONNX Export

from ragatouille.models.utils import export_to_huggingface_hub

# Upload model AND include a Vespa-compatible ONNX export
export_to_huggingface_hub(
    colbert_path="experiments/colbert/none/2024-01/checkpoints/colbert",
    huggingface_repo_name="myuser/my-colbert-model",
    export_vespa_onnx=True,
    use_tmp_dir=True,  # Re-save to temp dir for clean export
)

Related Pages

• Environment:AnswerDotAI_RAGatouille_Python_ColBERT_Dependencies