Platform Integrations (W&B, Hugging Face, Kaggle)¶

Abdelkader Haddag

Deep Learning Engineer & Researcher • 📅 Aug 5, 2026

This tutorial covers how to utilize the pre-built platform helpers in src/lib/integrations/ to connect training runs to Weights & Biases (W&B), upload model weights to Hugging Face Hub, and download datasets from Kaggle and Roboflow.

📊 Weights & Biases (W&B) Experiment Tracking¶

Enable real-time loss curves, system metrics, and artifact logging:

from src.lib.integrations.wandb import WandbIntegration

# Initialize W&B run
wandb_logger = WandbIntegration(
    project="deep-learning-boilerplate",
    name="experiment_resnet_v1",
    config={"learning_rate": 0.001, "batch_size": 64},
)

# Log training step metrics
wandb_logger.log_metrics({"train/loss": 0.245, "val/accuracy": 0.942}, step=epoch)

# Save checkpoint artifact
wandb_logger.log_artifact(
    name="model-checkpoint",
    type_name="model",
    filepath="outputs/checkpoints/best_model.pt",
)
wandb_logger.finish()

🤗 Hugging Face Hub Checkpoint Export¶

Upload trained PyTorch model checkpoints directly to the Hugging Face Model Hub:

from src.lib.integrations.huggingface import HuggingFaceIntegration

hf_helper = HuggingFaceIntegration(repo_id="your-username/my-resnet-model")

# Upload model checkpoint
hf_helper.upload_model(
    checkpoint_path="outputs/checkpoints/best_model.pt",
    commit_message="Upload trained ResNet baseline checkpoint",
)

🏆 Kaggle & Roboflow Dataset Automation¶

Download datasets directly into inputs/datasets/:

from src.lib.integrations.kaggle import KaggleIntegration
from src.lib.integrations.roboflow import RoboflowIntegration

# Download Kaggle competition dataset
kaggle = KaggleIntegration()
kaggle.download_dataset(dataset_name="cifar10", output_dir="inputs/datasets/cifar10")

# Download Roboflow object detection dataset
rf = RoboflowIntegration(api_key="YOUR_ROBOFLOW_KEY")
rf.download_dataset(
    workspace="vision-research",
    project="object-detection-v1",
    version=1,
    output_dir="inputs/datasets/roboflow_dataset",
)