NeuronScope
RecipeStablesince v0.6.0

Cloud storage

Read and write .nsz artifacts against S3, GCS, and Azure Blob with the same URI scheme.

Audience: Ops, ML engineersRead: 5 minEdit on GitHub

URI schemes

Every read/write path in NeuronScope accepts a URI. Local paths are the default; cloud backends are enabled by installing the matching extra.

·bash
pip install "neuronscope[s3]"
ns save run.nsz s3://team-runs/2026/07/run-1234.nsz

Credentials

NeuronScope does not manage credentials — it defers to each provider's standard chain (~/.aws/credentials, gcloud auth, DefaultAzureCredential). Use the platform's own tooling to rotate keys; nothing in the library caches them.

Throughput tips

  • Set NS_IO_CONCURRENCY=8 to parallelise chunk uploads for large artifacts.
  • Prefer a bucket in the same region as your training host — cross-region reads dominate wall time.
  • For CI, write to a local path and upload separately with the provider CLI; NeuronScope's uploader is convenient, not fast.
Related