Weights & Biases¶
fenn provides native integration with Weights & Biases (W&B) for comprehensive experiment monitoring. This integration automatically tracks metrics, hyperparameters, system stats, and model artifacts without requiring boilerplate code in your training scripts.
Prerequisites¶
To enable tracking, you must first authenticate the client:
- Log in to your W&B account.
- Navigate to User Settings > API Keys.
- Copy your personal API key.
Configuration¶
Enable the W&B integration by configuring your environment credentials and project settings.
Credentials (.env)¶
Store your API key securely in the .env file at the project root. fenn detects this variable automatically to authenticate the session.
WANDB_API_KEY=your_api_key_here
Project Settings (fenn.yaml)¶
Add a wandb block to your fenn.yaml configuration. The presence of this section signals fenn to initialize the logger.
wandb:
entity: your_wandb_username_or_team
Execution¶
Once configured, fenn manages the W&B lifecycle internally. When you execute your standard entry point:
python main.py
The framework automatically performs the following:
- Authentication: Loads the API key from the environment.
- Initialization: Calls wandb.init with the settings defined in fenn.yaml.
- Config Sync: Uploads the full experiment configuration (hyperparameters, architecture settings) to the W&B dashboard.
- Metric Streaming: Begins streaming loss curves, accuracy metrics, and logs in real-time.