For fast page loading in the W&B UI, we recommend keeping logged data amounts within these bounds.
Scalars: you can have tens of thousands of steps and hundreds of metrics
Histograms: we recommend limiting to thousands of steps
If you send us more that that, your data will be saved and tracked, but pages may load more slowly.
Generally you shouldn't be calling
wandb.log more than a few times per second or wandb may start to interfere with your training run's performance. We do not assert any limits beyond rate limiting. Our Python client will automatically do an exponential backoff and retry requests that exceed limits, so this should be transparent to you. It will say “Network failure” on the command line. For unpaid accounts, we may reach out in extreme cases where usage exceeds reasonable thresholds.
The W&B API is rate limited by IP and API key. New accounts are restricted to 200 requests per minute. This rate allows you to run approximately 15 processes in parallel and have them report without being throttled. If the wandb client detects it's being limited, it will backoff and retry sending the data in the future. If you need to run more than 15 processes in parallel send an email to email@example.com.
The maximum file size for new accounts is 2GB. A single run is allowed to store 10 GB of data. If you need to store larger files or more data per run, contact us at firstname.lastname@example.org.
Metrics are sampled to 1500 data points by default before displaying in the UI.
While a run is in progress we tail the last 5000 lines of your log for you in the UI. After a run is completed the entire log is archived and can be downloaded from an individual run page.
Here are some additional guidelines for logging data to W&B.
Nested parameters: We automatically flatten nested parameters, so if you pass us a dictionary we will turn it into a dot-separated name. For config values, we support 3 dots in the name. For summary values, we support 4 dots.