Wandb.jl

Unofficial Julia Bindings for wandb.ai.

To install simply do the following in a Julia REPL

] add Wandb

Follow the quickstart points 1 and 2 to get started with a Wandb account.

using Wandb, Logging

# Initialize the project
lg = WandbLogger(; project = "Wandb.jl", name = nothing)

# Set logger globally / in scope / in combination with other loggers
global_logger(lg)

# Logging Values
Wandb.log(lg, Dict("accuracy" => 0.9, "loss" => 0.3))

# Even more conveniently
@info "metrics" accuracy=0.9 loss=0.3
@debug "metrics" not_print=-1  # Will have to change debug level for this to be logged

# Tracking Hyperparameters
update_config!(lg, Dict("dropout" => 0.2))

# Close the logger
close(lg)

• Transition to use CondaPkg and PythonCall.
• save is no longer exported since the name is too common.
• Wandb.finish has been removed. Use Base.close instead.

• Base.log is not extended for WandbBackend of FluxTraining.

• Wandb.Image now supports any object with a show(::IO, ::MIME"image/png", img) (or image/jpeg) method.

• version() returns a VersionNumber instead of a String
• update_client() needs to be called to update the wandb client. We no longer check for updates by default.

• The Wandb python client version number is no longer printed during __init__ (i.e. when calling using Wandb). Instead, call Wandb.version() to see the client version number.

• Base.log is no longer exported. Users need to do Wandb.log (#9)
• FluxMPI + Wandb integration demo updated to the latest API