Dabus

A Julia package for visualizing Flux.jl neural network architectures as GraphViz diagrams.

---

Installation

This package is registered in LabRegistry

# With LabRegistry added in julia registry
] add Dabus

# Otherwise
] add https://github.com/lemieux-lab/Dabus

Or for local development:

] dev path/to/Dabus

---

Usage

using Dabus, Flux
 
model = Chain(
    Dense(50, 10, relu),
    Dense(10, 1)
)
 
# Display the network (returns image bytes)
img = draw_network(model)
 
# Save directly to a file
draw_network(model, save_to="network.png")
 
# Export as SVG
draw_network(model, save_to="network.svg", output_type="svg")

Note: Dabus requires an internet connection. Diagrams are rendered via the QuickChart.io GraphViz API : no local GraphViz installation needed.

---

Supported Layer Types

Category	Layers
Standard	Dense, Embedding
Convolutional	Conv, ConvTranspose, CrossCor
Pooling	MaxPool, MeanPool, AdaptiveMaxPool, AdaptiveMeanPool, GlobalMeanPool
Attention	MultiHeadAttention
Recurrent	LSTM, GRU
Containers	Chain, Parallel, Maxout, PairwiseFusion
Skip connections	SkipConnection
Activations / misc	Any callable (e.g. relu, softmax, Flux.flatten)

---

API Reference

draw_network(network; save_to=nothing, output_type="png")

Generates a diagram of a Flux neural network.

Arguments:

• network : A Flux model (e.g. a Chain or any supported layer).
• save_to : Optional file path. If provided, the image bytes are written to this path.
• output_type : Output format: "png" (default) or "svg".

Returns: Vector{UInt8} : the raw image bytes.

---

Examples

Embedding + Dense

Dense + softmax activation

Convolutional layers (Conv, ConvTranspose, CrossCor)

Pooling layers

Multi-head attention

LSTM recurrent network

Complex parallel architecture

---

Requirements

• Julia ≥ 1.0
• Flux.jl ≥ 0.16
• HTTP.jl ≥ 1.0
• Internet connection (diagrams are rendered via the QuickChart.io API)