leba0/nncase

Neural network optimization toolkit for fast inference

nncase is a neural network compiler for AI accelerators.

nncase 是一个为 AI 加速器设计的神经网络编译器。 技术交流 QQ 群：790699378

Install

安装

Download prebuilt binaries from Release.

下载预编译的二进制文件 Release。

Dependencies

依赖

gcc >= 8, cmake >=3.8, xtensor, xtl, clipp

---

Architecture

架构

Support commonly used CNN networks

支持常用的 CNN 网络

• MobileNetV1/V2
• YOLOV1 YOLOV3

Features

• Supports multiple inputs and outputs and multi-branch structure
• Static memory allocation, no heap memory acquired
• Operators fusion and optimizations
• Support float and quantized uint8 inference
• Support post quantization from float model with calibration dataset
• Flat model with zero copy loading

功能

• 支持多输入输出网络，支持多分支结构
• 静态内存分配，不需要堆内存
• 算子合并和优化
• 支持 float 和量化 uint8 推理
• 支持训练后量化，使用浮点模型和量化校准集
• 平坦模型，支持零拷贝加载

Usage

使用方法

Usage 使用方法

Examples 例子

Supported operators

支持的算子

Operator	Is Supported
Add	✅
ArgMax	❌
ArgMin	❌
AveragePool2D	✅
BatchToSpaceND	❌
Cast	❌
Concatenation	✅
Conv2D	✅
DepthwiseConv2D	✅
Div	✅
Equal	❌
Exp	✅
ExpandDims	❌
Floor	✅
FullyConnected	✅
Gather	❌
Greater	❌
GreaterEqual	❌
MaxPool2D	✅
Mean	✅
Mul	✅
L2Normalization	✅
L2Pool2D	❌
LessEqual	❌
Log	✅
Logistic	✅
LogSoftmax	❌
Maximum	✅
Minimum	✅
Neg	✅
NotEqual	❌
Pack	❌
Pad	✅
Pow	❌
PRelu	❌
ReduceMax	✅
ReduceProd	❌
Reshape	✅
ResizeBilinear	✅
Rsqrt	✅
Select	❌
Shape	❌
Sin	✅
Slice	❌
Softmax	✅
SpaceToDepth	❌
SpaceToBatchND	❌
SparseToDense	❌
Split	❌
Sqrt	✅
Square	✅
Squeeze	❌
Sub	✅
Sum	✅
Tile	❌
TopK	❌
Transpose	✅
TransposeConv	❌
LogicalOr	❌
OneHot	❌
LogicalAnd	❌
LogicalNot	❌
UnPack	❌
ReduceMin	✅
FloorDiv	❌
ReduceAny	❌
ZerosLike	❌
Fill	❌
FloorMod	❌
Range	❌
ResizeNearesetNeighbor	✅
LeakyRelu	✅
MirrorPad	❌
Abs	✅
SplitV	❌
Unique	❌
Ceil	✅
Reverse	❌
AddN	❌
GatherND	❌
Cos	✅
Where	❌
Rank	❌
Elu	❌
ReverseSequence	❌