English
sofia-engine
edge-ai
industrial-ai
scientific-computing
embedded-ai
signal-processing
digital-signal-processing
predictive-maintenance
condition-monitoring
vibration-analysis
anomaly-detection
industrial-iot
iiot
telemetry
edge-computing
tinyml
on-device-learning
embedded-systems
machine-health
time-series
python
typescript
c
| # Sofia Engine Checkpoint & Artifact Registry | |
| This directory defines the distribution and security policy for future trained Sofia model checkpoints and parameter sets. | |
| --- | |
| ## 1. Current Checkpoint Status | |
| > **Canonical Pretrained Checkpoint:** *Not yet released.* | |
| > | |
| > Sofia Engine 3.0.0a1 provides the runtime engine, feature extraction contracts, assembly virtual machine, and safe serialization interfaces. No pre-trained foundation model weights are shipped in this release. | |
| --- | |
| ## 2. Checkpoint Release Policy | |
| Every future checkpoint published under this repository must satisfy the following technical requirements: | |
| 1. **Safe Serialization Formats**: | |
| - Primary: `.safetensors` or validated `.npz` (via `np.savez`). | |
| - **Prohibited:** Python `pickle` (`.pkl`, `.pickle`, `.pt` containing arbitrary executable code) is strictly forbidden. | |
| 2. **Cryptographic Integrity**: | |
| - Every artifact file must be accompanied by its lowercase SHA-256 digest (`parameters_sha256`). | |
| - Manifests are verified before weights are mapped into memory. | |
| 3. **Declared Feature Schema**: | |
| - The checkpoint manifest must explicitly declare its input `feature_schema` (e.g., `vibration-v3`, `power-quality-v1`) to prevent feature order mismatches. | |
| 4. **Transparent Provenance**: | |
| - Public release notes must specify training dataset origin, sample count, evaluation metrics, and runtime version compatibility. | |