Skip to content

neo-ai/neo-loader

Neo Framework -> Relay IR converters

This is a set of utility classes for converting ML models to TVM Relay IR.

Installation

pip install NeoCompilerModelLoaders

This package does not assume any framework dependencies. It is the responsibility of the consumer to have installed the framework versions that they choose to support.

Usage

import tarfile

from neo_loader import load_model

model_artifacts = []
with tarfile.open('/path/to/model.tar.gz', 'r:gz') as tf:
    tf.extractall()
    model_artifacts = tf.getnames()

relay_artifacts = load_model(
    model_artifacts=model_artifacts,
    input_shape={'data': [1, 3, 224, 224]},
    framework='tensorflow'
)

def relay_func(relay_artifacts: Dict) -> object:
    return relay_artifacts['model_objects'][0]

def relay_params(relay_artifacts: Dict) -> object:
    return relay_artifacts['model_objects'][1]

def relay_dtype(relay_artifacts: Dict) -> object:
    if len(relay_artifacts['model_objects']) == 2:
        return {}
    else:
        return relay_artifacts['model_objects'][2]

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.

About

No description, website, or topics provided.

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages