Crate jugar_apr 

Aprender Package Resource (.apr) model format.

Per spec Section 4.1: The .apr format is a compact, portable container for trained AI behaviors that can be hot-swapped like trading cards.

File Structure

┌─────────────────────────────────────────────────────────────┐
│                    .APR File Structure                       │
├─────────────────────────────────────────────────────────────┤
│  Magic Number: "APNR" (4 bytes)                             │
│  Version: u16 (2 bytes)                                      │
│  Checksum: CRC32 (4 bytes)                                   │
├─────────────────────────────────────────────────────────────┤
│  Metadata (CBOR encoded):                                    │
│    - name, version, author, license                          │
│    - difficulty_levels, input_schema, output_schema          │
├─────────────────────────────────────────────────────────────┤
│  Model Data (compressed):                                    │
│    - weights: [f32; N]                                       │
│    - biases: [f32; M]                                        │
│    - architecture: string                                    │
└─────────────────────────────────────────────────────────────┘

Structs

AprFile

Parsed APR file

AprMetadata

Metadata for an APR model

AprModel

A complete APR model with metadata and data

ModelData

Neural network weights and architecture

Enums

AprError

Errors that can occur when working with APR models

ModelArchitecture

Network architecture specification

Constants

APR_MAGIC

Magic number for APR files

APR_VERSION

Current APR version

CURRENT_VERSION

Current model version

MAX_MODEL_SIZE

Maximum allowed model size (1 MB per spec Section 9.1)

MIN_SUPPORTED_VERSION

Minimum model version supported