Backblaze B2 sample notebooks

Example notebooks demonstrating how to use Backblaze B2 Cloud Storage with AI and data workflows. Each subdirectory is a self-contained example with its own README.md, dependencies, and runnable notebook.

Examples

image-classification-pytorch/

Train a PyTorch CIFAR-10 image classifier on data hosted in Backblaze B2. Demonstrates a custom PyTorch Dataset that streams training images from a B2 bucket via the S3-compatible API, learning to recognize the 10 CIFAR-10 categories (airplanes, cars, birds, cats, deer, dogs, frogs, horses, ships, trucks).

ray-train-tune-checkpoints/

End-to-end Ray Train + Ray Tune example with checkpoints on Backblaze B2. Reads California Housing parquet from a public B2 bucket, trains a small PyTorch regression model with TorchTrainer, writes checkpoints back to a private B2 bucket via RunConfig(storage_path="s3://..."), and runs an optional Tuner sweep over learning rates. Companion to the Ray Train persistent-storage user guide.

whisper-b2-transcription/

Speech-to-text transcription on Backblaze B2 with OpenAI Whisper. Streams a public-domain demo audio clip (jfk.flac) from a public B2 bucket via the S3-compatible API, runs Whisper for ASR, and optionally writes the transcript JSON back to a private B2 bucket. Starting point for batch transcription pipelines on B2-hosted audio archives.

How to run an example

Each example directory has its own README.md with detailed setup instructions, but in short:

• In a browser, click one of the launch badges (Colab, Binder, Codespaces) on the example you want.
• Locally, clone this repo, cd into the example directory, and follow its README.md (typically pip install -r requirements.txt && jupyter lab <notebook>.ipynb).

Backblaze B2 prerequisites

Most examples need a Backblaze B2 application key. Generate one at https://www.backblaze.com/docs/cloud-storage-application-keys, then export the values as the standard AWS-named environment variables (B2's S3-compatible API reads these under the AWS SDK):

export AWS_ENDPOINT_URL_S3="https://s3.<region>.backblazeb2.com"  # region from B2 console
export AWS_ACCESS_KEY_ID="<your B2 application key ID>"
export AWS_SECRET_ACCESS_KEY="<your B2 application key>"

For Colab / Codespaces / Kaggle / Binder, see the per-example README.md for the secret-store path that fits each runtime.

Contributing

New examples are welcome. Each example lives in its own top-level directory and is expected to include:

• A descriptive README.md (purpose, how to run, secret setup if needed)
• The notebook(s) themselves, with launch badges that point at the path on main
• A requirements.txt (or equivalent) so the example is reproducible
• A .github/workflows/test-<name>.yml workflow that executes the notebook end-to-end against the shared backblaze-samples-ci bucket

See CLAUDE.md for the full conventions: writing style (no em dashes), repo layout, two-bucket pattern, custom user_agent_extra on every B2 boto3 client, headless-execution env vars, pre-commit hooks, and the per-notebook CI workflow shape.

Before opening a PR, install and run pre-commit locally:

pip install pre-commit
pre-commit install
pre-commit run --all-files

CI runs the same hooks via .github/workflows/lint.yml on every push.

Related

• Backblaze B2 Cloud Storage
• Backblaze developer docs
• More B2 sample apps across languages and frameworks: https://github.com/backblaze-b2-samples