Module serialization 

Model Serialization Module

Implements SafeTensors format for model serialization, compatible with:

• realizar inference engine
• HuggingFace ecosystem
• Ollama (can convert to GGUF)
• PyTorch, TensorFlow

SafeTensors Format:

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

Example:

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();

Re-exports

pub use safetensors::SafeTensorsMetadata;

Modules

safetensors

SafeTensors format implementation for model serialization.