creyone-layer 1.0.5

PyTorch layer building tools for CREYONE

creyone_layer (Beta Package for CreYoNe)

Create Your Network (a.k.a. CreYoNe) is a utility tools for making DNN module via torch. This repository provides Layer-sized building blocks for deep learning.

Installation

pip install creyone-layer

From source:

git clone https://github.com/qnilix/creyone_layer.git
cd creyone_layer
pip install -e .

Quick Start

Layer registry — create_layer

from creyone_layer import create_layer

relu = create_layer('relu', 'act')(inplace=True)()
bn   = create_layer('batch', 'norm')(dim=2, eps=1e-5, mom=0.1)(64)
conv = create_layer('base',  'conv')(dim=2, optional='ap')(32, 64, 3)
pool = create_layer('max',   'pool')(dim=2, optional='ap')(2)

Registered Layers

Family	Names
conv	base, depthwise
norm	batch, layer
act	relu, relu6, gelu, quickgelu, sigmoid, silu, hardsig, hardswish, swisheff
pool	max, avg

conv options (passed via optional='...', +-separated)

Flag	Effect
ap	auto-pad — output spatial size matches input
dw	depthwise — groups = in_channels
grid	stride = kernel size (grid-like sampling)
ar	AutoReshape — accepts (B, H*W, C) token sequences