Hugging Face Transformers provides general-purpose architectures for Natural Language Understanding (NLU) and Natural Language Generation (NLG) with pretrained models in 100+ languages and deep interoperability between TensorFlow 2.0 and PyTorch.
To get training logged automatically, just install the library and log in:
pip install wandbwandb login
The Trainer
or TFTrainer
will automatically log losses, evaluation metrics, model topology and gradients.
Advanced configuration is possible through wandb environment variables.
Additional variables are available with transformers:
Environment Variables | Options |
WANDB_WATCH |
|
WANDB_DISABLED | boolean: Set to true to disable logging entirely |
We've created a few examples for you to see how the integration works:
Run in colab: A simple notebook example to get you started
A step by step guide: track your Hugging Face model performance
Does model size matter? A comparison of BERT and DistilBERT
We'd love to hear feedback and we're excited to improve this integration. Contact us with any questions or suggestions.
Explore your results dynamically in the W&B Dashboard. It's easy to look across dozens of experiments, zoom in on interesting findings, and visualize highly dimensional data.
Here's an example comparing BERT vs DistilBERT — it's easy to see how different architectures effect the evaluation accuracy throughout training with automatic line plot visualizations.