Launch Experiments with wandb.init
Call wandb.init() at the top of your script to start a new run
Call
wandb.init()
once at the beginning of your script to initialize a new job. This creates a new run in W&B and launches a background process to sync data.- Automated Environments: Most of these settings can also be controlled via Environment Variables. This is often useful when you're running jobs on a cluster.
View the reference docs for this function, generated from the
wandb
Python library.If you're trying to start multiple runs from one script, add two things to your code:
- 1.
run = wandb.init(reinit=True)
: Use this setting to allow reinitializing runs - 2.
run.finish()
: Use this at the end of your run to finish logging for that run
import wandb
for x in range(10):
run = wandb.init(reinit=True)
for y in range (100):
wandb.log({"metric": x+y})
run.finish()
Alternatively you can use a python context manager which will automatically finish logging:
import wandb
for x in range(10):
run = wandb.init(reinit=True)
with run:
for y in range(100):
run.log({"metric": x+y})
This error indicates that the library is having difficulty launching the process which synchronizes data to the server.
The following workarounds can help resolve the issue in certain environments:
Linux / OS X
Google Colab
wandb.init(settings=wandb.Settings(start_method="fork"))
For versions prior to
0.13.0
we suggest using:wandb.init(settings=wandb.Settings(start_method="thread"))
If your training program uses multiple processes you will need to structure your program to avoid making wandb method calls from processes where you did not run
wandb.init()
.
There are several approaches to managing multiprocess training:- 1.Call
wandb.init
in all your processes, using the group keyword argument to define a shared group. Each process will have its own wandb run and the UI will group the training processes together. - 2.
Check out the Distributed Training Guide for more detail on these two approaches, including code examples with Torch DDP.
import wandb
wandb.init()
run_name = wandb.run.name
If you'd like to overwrite the run name (like snowy-owl-10) with the run ID (like qvlp96vk) you can use this snippet:
import wandb
wandb.init()
wandb.run.name = wandb.run.id
If you do not explicitly name your run, a random run name will be assigned to the run to help identify the run in the UI. For instance, random run names will look like "pleasant-flower-4" or "misunderstood-glade-2".
When
wandb.init
is called in your script, we automatically look for git information to save, including a link to a remote repo and the SHA of the latest commit. The git information should show up on your run page. If you aren't seeing it appear there, make sure that your shell's current working directory when executing your script is located in a folder managed by git.The git commit and command used to run the experiment are visible to you but are hidden to external users, so if you have a public project, these details will remain private.
By default,
wandb.init
starts a process that syncs metrics in real time to our cloud hosted app. If your machine is offline, you don't have internet access, or you just want to hold off on the upload, here's how to run wandb
in offline mode and sync later.- 1.
- 2.
WANDB_MODE="offline"
And here's a sample of what this would look like in your script:
import wandb
import os
os.environ["WANDB_API_KEY"] = YOUR_KEY_HERE
os.environ["WANDB_MODE"] = "offline"
config = {
"dataset": "CIFAR10",
"machine": "offline cluster",
"model": "CNN",
"learning_rate": 0.01,
"batch_size": 128,
}
wandb.init(project="offline-demo")
for i in range(100):
wandb.log({"accuracy": i})
Here's a sample terminal output:

And once you're ready, just run a sync command to send that folder to the cloud.
wandb sync wandb/dryrun-folder-name

Modes can be "online", "offline" or "disabled", and default to online.
online
(default): In this mode, the client sends data to the wandb server.offline
: In this mode, instead of sending data to the wandb server, the client will store data on your local machine which can be later synced with the wandb sync
command.disabled
: In this mode, the client **** returns mocked objects and prevents all network communication. The client will essentially act like a no-op. In other words, all logging is entirely disabled. However, stubs out of all the API methods are still callable. This is usually used in tests.You most likely lost connection to your machine while training. You can recover your data by running
wandb sync [PATH_TO_RUN]
. The path to your run will be a folder in your wandb
directory corresponding to the Run ID of the run in progress.If you're getting the error message
Launch Error: Permission denied
, you don't have permissions to log to the project you're trying to send runs to. This might be for a few different reasons.- 1.
- 2.You've set an entity that doesn't exist. "Entity" should be your username or the name of an existing team. If you need to create a team, go to our Subscriptions page.
- 3.You don't have project permissions. Ask the creator of the project to set the privacy to Open so you can log runs to this project.
Yes, W&B uses the
multiprocessing
library. If you see an error message such as:An attempt has been made to start a new process before the current process
has finished its bootstrapping phase.
This might mean that you might need to add an entry point protection
if name == main
. Note that you would only need to add this entry point protection in case you're trying to run W&B directly from the script.Last modified 1mo ago