Skip to content

Model Upload & Download

Installation

Install infernet-ml from your terminal:

uv pip install infernet-ml

pip install infernet-ml

Repository Id's

Ritual's artifacts are stored in repositories. These repositories are identified by a unique id. The id is a string that is formatted as follows: storage/username/repo-name.

For example, the following are valid repository id's:

  • arweave/my-username/my-model
  • huggingface/my-username/my-model

Each component of a repository id is as follows:

  • storage: The storage provider. This can be either arweave or huggingface.
  • username: The username of the user who owns the repository.
  • repo-name: The name of the repository.

For more information refer to RitualRepoId's reference.

RitualRepoId provides utility methods to convert a repository id to a RitualRepoId object.

id = RitualRepoId.from_unique_id("arweave/my-username/my-model")

You can also convert a RitualRepoId object to a unique id.

id = RitualRepoId("arweave", "my-username", "my-model")
print(id.to_unique_id())

Models are uploaded to Repositories, so for the next sections, we will refer to the repository id as repo_id.

Uploading Models

Infernet 2.0 adds a utility class called ModelManager. This facilitates the uploading and downloading of models to various data sources.

Arweave Upload

Put your model & relevant files in a directory and run the following code. You'll have to pass in the directory of the model, the repo_id of the model, and the metadata.

from infernet_ml.utils.model_manager import ModelManager

model_manager = ModelManager()
directory = "path/to/model"
repo_id = "arweave/my-username/my-model"

ModelManager.upload_model(
    directory=directory,
    repo_id=repo_id,
    metadata={
        "my-metadata-key": "my-metadata-value",
    },
    wallet_path="./my-wallet.json",
)

The metadata parameter is a dictionary that will be stored with the model. The wallet_path parameter is the path to your Arweave wallet. This wallet will be used to pay for the transaction fees associated with uploading the model.

Huggingface Upload

Uploading to huggingface is the same as arewave, except instead of wallet_path, you'll need to pass in a huggingface token (hf_token).

from infernet_ml.utils.model_manager import ModelManager

model_manager = ModelManager()
directory = "path/to/model"
repo_id = "huggingface/my-username/my-model"

ModelManager.upload_model(
    directory=directory,
    repo_id=repo_id,
    metadata={},
    hf_token="my-huggingface-token",
)

Downloading Models

To download a model, you'll need to know the repo_id of the model you want to download, as well as the specific file-name of the model you want to download. The repo_id as well as the file_name together will comprise the model_id. The format of the model_id is repo_id:file_name.

If your model has multiple files, you can comma-separate the file names i.e. repo_id:file1.onnx,file2.onnx.

Arweave Download

from infernet_ml.utils.model_manager import ModelManager

model_manager = ModelManager()
repo_id = "arweave/my-username/my-model"
file_name = "model.onnx"
model_id = f"{repo_id}:{file_name}"

ModelManager.download_model(
    repo_id=repo_id,
    file_name=file_name,
    output_dir="path/to/output",
)

Huggingface Download

If your model is private, you'll need to pass in a huggingface token (hf_token).

from infernet_ml.utils.model_manager import ModelManager

model_manager = ModelManager()
repo_id = "huggingface/my-username/my-model"
file_name = "model.onnx"
model_id = f"{repo_id}:{file_name}"

ModelManager.download_model(
    repo_id=repo_id,
    file_name=file_name,
    output_dir="path/to/output",
    hf_token=hf_token,
)