This library provides an abstraction to perform Model Versioning using Weight & Biases.

Last update: Jan 28, 2022

Overview

Description

This library provides an abstraction to perform Model Versioning using Weight & Biases.

Features

Version a new trained model
Promote a model to another stage (e.g production)

Example of usage

The following code snippet shows how to promote a newly trained model to the best_model on the validation set. First, it creates the model providing the checkpoint path, the artifact name and type, a description, and the metadata, where the metrics or any desired information is stored.

The comparision_type indicates how to compare the metrics between the two models. E.g if smaller, the smaller the metric from the new model the better

run = wandb.init(...)

# Create the desired artifact
artifact = versioner.create_artifact(
                            checkpoint='model.ckpt',
                            artifact_name='prueba',
                            artifact_type='model',
                            description='Prueba Wandb-MV',
                            metadata={
                                'val_metric': 78.0,
                                'test_metric': 0.0
                            })

# Promote the desired artifact to the 'best_model' tag
versioner.promote_model(new_model=artifact,
                        artifact_name='prueba',
                        artifact_type='model',
                        comparision_metric='val_metric',
                        promotion_alias='best_model',
                        comparision_type='smaller'
                       )

This code snippet shows how to promote a trained model to production after being validated on the test set. The already_deployed parameter indicates that the model is already logged, so it only needs to be updated.

versioner = Versioner(run)
versioner.promote_model(new_model=artifact,
                        artifact_name='prueba',
                        artifact_type='model',
                        comparision_metric='test_metric',
                        promotion_alias='production',
                        comparision_type='smaller',
                        already_deployed=True
                        )

Next features

Allow providing a custom comparision function to promote a model
...

This library provides an abstraction to perform Model Versioning using Weight & Biases.

Related tags

Overview

Description

Features

Example of usage

Next features

Owner

Hector Lopez Almazan

Official implementation of VQ-Diffusion

Nvdiffrast - Modular Primitives for High-Performance Differentiable Rendering

Cooperative Driving Dataset: a dataset for multi-agent driving scenarios

Pytorch implementation of Learning with Opponent-Learning Awareness

A pytorch implementation of Reading Wikipedia to Answer Open-Domain Questions.

Contextualized Perturbation for Textual Adversarial Attack, NAACL 2021

Rainbow DQN implementation that outperforms the paper's results on 40% of games using 20x less data 🌈

Repo for 2021 SDD assessment task 2, by Felix, Anna, and James.

Deep generative modeling for time-stamped heterogeneous data, enabling high-fidelity models for a large variety of spatio-temporal domains.

Fuse radar and camera for detection

Explainability of the Implications of Supervised and Unsupervised Face Image Quality Estimations Through Activation Map Variation Analyses in Face Recognition Models

Reference PyTorch implementation of "End-to-end optimized image compression with competition of prior distributions"

Simple Baselines for Human Pose Estimation and Tracking

Pytorch implementation of the paper "Class-Balanced Loss Based on Effective Number of Samples"

TextureGAN in Pytorch

A Weakly Supervised Amodal Segmenter with Boundary Uncertainty Estimation

PyTorch reimplementation of the Smooth ReLU activation function proposed in the paper "Real World Large Scale Recommendation Systems Reproducibility and Smooth Activations" [arXiv 2022].

基于深度强化学习的原神自动钓鱼AI

Few-NERD: Not Only a Few-shot NER Dataset

A pure PyTorch batched computation implementation of "CIF: Continuous Integrate-and-Fire for End-to-End Speech Recognition"