Module serialization 

Model Serialization Module

Provides two serialization formats:

SafeTensors Format

Industry-standard format compatible with HuggingFace ecosystem.

[8-byte header: u64 metadata length (little-endian)]
[JSON metadata: tensor names, dtypes, shapes, data_offsets]
[Raw tensor data: F32 values in little-endian]

APR Format

Compact binary format optimized for WASM deployment with JSON metadata.

[4-byte magic: "APR1"]
[4-byte metadata_len][JSON metadata]
[4-byte n_tensors][tensor index][tensor data]
[4-byte CRC32]

Example (SafeTensors):

use aprender::linear_model::LinearRegression;
use aprender::primitives::{Matrix, Vector};
use aprender::traits::Estimator;

let mut model = LinearRegression::new();
model.fit(&x, &y).unwrap();

// Save to SafeTensors
model.save_safetensors("model.safetensors").unwrap();

// Load from SafeTensors
let loaded = LinearRegression::load_safetensors("model.safetensors").unwrap();

Example (APR with metadata):

use aprender::serialization::apr::{AprWriter, AprReader};
use serde_json::json;

let mut writer = AprWriter::new();
writer.set_metadata("model_type", json!("whisper-tiny"));
writer.set_metadata("vocab", json!(["hello", "world"]));
writer.add_tensor_f32("weights", vec![2, 2], &[1.0, 2.0, 3.0, 4.0]);

let bytes = writer.to_bytes().unwrap();
let reader = AprReader::from_bytes(bytes).unwrap();
assert_eq!(reader.get_metadata("model_type").unwrap(), "whisper-tiny");

Re-exports

pub use apr::AprReader;

pub use apr::AprWriter;

pub use safetensors::SafeTensorsMetadata;

Modules

apr

APR (Aprender) binary format with JSON metadata.

safetensors

SafeTensors format implementation for model serialization.