sensorlm-rs 0.1.0

SensorLM – wearable sensor foundation model in Rust (Burn + WGPU)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

//! Model quantisation.
//!
//! | Module | Contents |
//! |--------|----------|
//! | [`int8`] | INT8 post-training quantisation, FP16 export, calibration |
//!
//! # Supported schemes
//!
//! | Scheme | Memory saving | Quality loss |
//! |--------|---------------|--------------|
//! | FP16 (half precision) | 2× | Negligible |
//! | INT8 symmetric per-tensor | 4× | Small (~0.5–1 % on retrieval tasks) |
//! | INT8 symmetric per-channel | 4× | Very small (~0.1–0.3 %) |
//!
//! # Workflow
//!
//! ```no_run
//! use sensorlm::quantization::int8::{quantize_model_weights, QuantizedModel};
//! use std::path::Path;
//!
//! // Extract weights from a trained model (pseudo-code):
//! // let weights = extract_linear_weights(&trained_model);
//! // let qm = quantize_model_weights(config_json, weights.into_iter());
//! // qm.save(Path::new("model_int8.json")).unwrap();
//! ```

pub mod int8;