Distributed storage for ML checkpoints: shards .safetensors across a Raspberry Pi cluster over TCP with SHA-256 verification, replication and zero-config mDNS discovery.
-
Updated
Oct 7, 2026 - Python
Distributed storage for ML checkpoints: shards .safetensors across a Raspberry Pi cluster over TCP with SHA-256 verification, replication and zero-config mDNS discovery.
Portable model checkpoint metadata and offline SHA-256 integrity verification — no W&B, MLflow, or tracking server.
GRIT — Grouped Reduced-precision Interchange Type: a checkable 64-byte descriptor + contract for block-scaled, sparse, sub-byte tensors. Spec, conformance vectors, and 5 zero-dependency implementations (C, C++, Rust, Python, TypeScript). grit scan audits GGUF/safetensors files you already have.
To associate your repository with the model-checkpoints topic, visit your repo's landing page and select "manage topics."