Can I just set the run name to the run ID?
Yes, set run.name = run.id after wandb.init() to use the run ID as the display name.
Can I log metrics on two different time scales?
Yes, log step indices like batch and epoch alongside metrics, then use define_metric to set the x-axis.
Can I run wandb offline?
Yes, set WANDB_MODE=offline to log locally, then use wandb sync to upload results later.
Does W&B use the `multiprocessing` library?
Yes, W&B uses multiprocessing. Add if name == ‘main’ to avoid bootstrapping errors.
Does logging block my training?
No, wandb.log writes to a local file and a separate process syncs data asynchronously.
Does your tool track or store training data?
W&B does not store training data unless you explicitly call wandb.Run.save() with a file.
How can I access the data logged to my runs directly and programmatically?
Use the W&B API to fetch run history with api.run(‘username/project/run_id’).history().
How can I compare images or media across epochs or steps?
Use the image panel’s step slider to compare model outputs across different training steps.
How can I configure the name of the run in my training code?
Set a custom run name by passing the name parameter to wandb.init in your training script.
How can I define the local location for `wandb` files?
Use environment variables like WANDB_DIR, WANDB_ARTIFACT_DIR, and others to control where W&B stores local files.
How can I organize my logged charts and media in the W&B UI?
Use the / character in metric names to group logged charts into panel sections in the W&B UI.
How can I save the git commit associated with my run?
W&B automatically saves the git remote URL and latest commit SHA when wandb.init is called from a git repo.
How can I see files that do not appear in the Files tab?
The Files tab shows up to 10,000 files. Use the public API to download all files from a run.
How can I use wandb with multiprocessing, e.g. distributed training?
Use grouped runs or single-process logging with multiprocessing queues for distributed training with W&B.
How do I add Plotly or Bokeh Charts into Tables?
Export Plotly or Bokeh figures to HTML and log them as wandb.Html objects in W&B Tables.
How do I fix `Rate limit exceeded` errors when logging metrics?
Reduce logging frequency, batch metrics, or upgrade your plan to resolve 429 rate limit errors.
How do I get the random run name in my script?
Access the auto-generated run name using the run object’s name attribute after calling wandb.init.
How do I launch multiple runs from one script?
Use wandb.init as a context manager to finish each run before starting the next one in a single script.
How do I log a list of values?
Log a list of values using wandb.log with dictionaries or wandb.Table for structured data.
How do I plot multiple lines on a plot with a legend?
Use wandb.plot.line_series() with the keys argument to create multi-line charts with a legend.
How do I programmatically access the human-readable run name?
Access the human-readable run name using the run.name attribute after calling wandb.init().
I didn't name my run. Where is the run name coming from?
W&B assigns a random two-word name like ‘pleasant-flower-4’ when you don’t set a run name.
InitStartError: Error communicating with wandb process
Resolve this error by changing the start_method setting to ‘fork’ or ‘thread’ for your platform.
Is it possible to save metrics offline and sync them to W&B later?
Set WANDB_MODE=offline to log metrics locally, then use wandb sync to upload them later.
My run's state is `crashed` on the UI but is still running on my machine. What do I do to get my data back?
Use wandb sync to recover data from a run that shows as crashed due to a lost connection.
What does wandb.init do to my training process?
wandb.init creates a run on the server and starts a background process to stream metrics.
What happens if I pass a class attribute into wandb.Run.log()?
Use a deep copy when logging class attributes to prevent values from changing before upload.
What happens when I log millions of steps to W&B? How is that rendered in the browser?
W&B samples data down to 1,500 points for rendering. Keep metrics under 10,000 points for best performance.
What if I want to integrate W&B into my project, but I don't want to upload any images or media?
W&B supports scalar-only logging. You can specify exactly which data to upload.
What if I want to log some metrics on batches and some metrics only on epochs?
Log a custom x-axis value alongside your metrics and select it in the chart editor.
What is the difference between wandb.init modes?
Learn the differences between online, offline, and disabled modes for wandb.init.
Why am I seeing fewer data points than I logged?
Metrics must be logged at the same step to stay synchronized. Bundle metrics into a single log call.
Why are steps missing from a CSV metric export?
CSV exports have row limits. Download the full run history as a Parquet artifact instead.
Why can't I sort or filter metrics with certain characters?
Metric names must follow GraphQL naming rules: letters, digits, and underscores only.
Why does my process hang when using Hydra with W&B?
Fix Hydra and W&B multiprocessing conflicts by setting the start method to thread.
Why does my training hang with distributed training?
Fix training hangs at the start or end of distributed training by enabling W&B Service and calling wandb.finish().
Why is nothing showing up in my graphs?
Graphs stay empty until the first wandb.log call executes. Log more frequently to see data sooner.
Why is the same metric appearing more than once?
Logging different data types under the same key causes duplicate metric entries. Use one type per key.
Will wandb slow down my training?
W&B has minimal performance impact under normal usage, running in a separate non-blocking process.