embedded-nn 0.2.0

A pure Rust, #![no_std] neural network inference library inspired by CMSIS-NN for microcontrollers and embedded targets.
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
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58

//! # embedded-nn
//!
//! `embedded-nn` is a pure Rust, `#![no_std]` neural network inference library for microcontrollers and embedded targets,
//! inspired by ARM's CMSIS-NN and TensorFlow Lite Micro.
//!
//! ## Modules
//! - [`types`]: Dimensional shapes, parameter structures, error types.
//! - [`support`]: Fixed-point quantization math, rounding division, bit operations.
//! - [`activations`]: ReLU, ReLU6, LeakyReLU, Sigmoid, Tanh.
//! - [`basic_math`]: Elementwise addition, subtraction, and multiplication.
//! - [`convolution`]: 2D Convolution, 1x1 Convolution, Depthwise Convolution, Transposed Conv, 1D Temporal Conv.
//! - [`fully_connected`]: Fully Connected (Linear / Dense) layers and Batch Matrix Multiplication (`BatchMatMul`).
//! - [`pooling`]: Max Pooling, Average Pooling.
//! - [`softmax`]: Softmax activation.
//! - [`concat`]: Depthwise concatenation.
//! - [`pad`]: Tensor padding.
//! - [`transpose`]: Matrix and spatial transposition.
//! - [`reshape`]: Reshaping operations.
//! - [`simd`]: Target SIMD vectorization abstractions, ARM DSP SMLAD assembly, and dot-product acceleration.
//! - [`subbyte`]: Sub-byte 4-bit (`s4`) quantization layers and packing helpers.
//! - [`recurrent`]: Recurrent neural network layers (LSTM cell, SVDF filter with 8-bit & 16-bit state).
//! - [`float_ops`]: Floating-point (`f32` & IEEE-754 `f16`) fallback layers.

#![no_std]
#![deny(missing_docs)]

pub mod activations;
pub mod basic_math;
pub mod concat;
pub mod convolution;
pub mod float_ops;
pub mod fully_connected;
pub mod pad;
pub mod pooling;
pub mod recurrent;
pub mod reshape;
pub mod simd;
pub mod softmax;
pub mod subbyte;
pub mod support;
pub mod transpose;
pub mod types;

pub use support::{
    clamp, divide_by_power_of_two, doubling_high_mult_no_sat, pack_q15x2_32x1, pack_s8x4_32x1,
    requantize, requantize_s64,
};
pub use types::{
    Activation, Context, ConvParams, Dims, DwConvParams, Error, FcParams, PerChannelQuantParams,
    PerTensorQuantParams, PoolParams, QuantParams, Result, SoftmaxParams, Tile,
};

pub use convolution::{convolve_1_x_n_s8, transpose_conv_s8};
pub use float_ops::{f16_to_f32, f32_to_f16};
pub use fully_connected::{batch_matmul_s16, batch_matmul_s8};
pub use recurrent::{lstm_step_s16, lstm_step_s8_s16, svdf_s8, svdf_state_s16_s8, LstmGateParams};
pub use simd::{vec_dot_s16, vec_dot_s8};
pub use subbyte::{convolve_s4, fully_connected_s4, pack_s4_pair, unpack_s4_pair};