MMF

Integrate W&B with Meta AI’s MMF framework to track multimodal model training experiments and log metrics.

This page describes how to use the WandbLogger class in Meta AI’s MMF library to track your multimodal model training with W&B. Enabling WandbLogger lets you log training and validation metrics, system (GPU and CPU) metrics, model checkpoints, and configuration parameters, so you can monitor experiments and compare runs without adding custom logging code.

Features

The WandbLogger in MMF supports the following features:

• Training and validation metrics
• Learning rate over time
• Model checkpoint saving to W&B Artifacts
• GPU and CPU system metrics
• Training configuration parameters

Configuration parameters

To turn on W&B logging and customize how runs are tracked, set the following options in your MMF configuration:

training:
    wandb:
        enabled: true
        
        # An entity is a username or team name where you're sending runs.
        # By default, it logs the run to your user account.
        entity: null
        
        # Project name to be used while logging the experiment with wandb
        project: mmf
        
        # Experiment/ run name to be used while logging the experiment
        # under the project with wandb. The default experiment name
        # is: ${training.experiment_name}
        name: ${training.experiment_name}
        
        # Turn on model checkpointing, saving checkpoints to W&B Artifacts
        log_model_checkpoint: true
        
        # Additional argument values that you want to pass to wandb.init() such as:
        # job_type: 'train'
        # tags: ['tag1', 'tag2']
        
env:
    # To change the path to the directory where wandb metadata is 
    # stored (Default: env.log_dir):
    wandb_logdir: ${env:MMF_WANDB_LOGDIR,}