Keep pulling the thread on Weights & Biases Models.
The Weights & Biases platform tracks model and dataset lineages to ensure reproducibility and audibility of ML projects.
For each experiment, Weights & Biases tracks the associated Git repository, commit hash, specific script used, and all parameters passed at training time.
Weights & Biases tracks model and data artifacts and provides a 'lineage view' to show the relationship between input and output artifacts and where they have been consumed.
Weights & Biases Automations can trigger webhook-based workflows in response to events like a new model version being linked to a registry or a new alias being added.
Weights & Biases Models is a platform designed to serve as a system of record for model training.
The Weights & Biases platform can track the end-to-end machine learning development cycle, including metrics from pre-training to evaluation and any generated artifacts.
Weights & Biases provides customizable workspaces and report generation tools to speed up debugging and collaboration.
The Weights & Biases platform can be run on any hardware, including public cloud environments or locally hosted infrastructure.
Weights & Biases is compatible with ML libraries including Keras, PyTorch, and Hugging Face Transformers.
The Weights & Biases platform can track hardware metrics such as GPU utilization and allocated memory for ML projects.
Weights & Biases allows users to set alerts on any tracked metric, which can be configured to notify specific users or channels.
The Weights & Biases platform tracks media associated with experiments, such as videos of simulations, to provide visual feedback on model performance at different training steps.