Why are my metrics missing from wandb.log()?

If metrics logged with wandb.log() are not appearing in the W&B UI, there are several common causes.

SDK version issue

Certain SDK versions have known bugs that cause logged metrics to be silently dropped. If you recently upgraded or installed W&B and are seeing missing metrics, check your SDK version:

pip show wandb

Update to the latest stable release:

pip install --upgrade wandb

If the issue started after a specific upgrade, you can pin to an earlier version while the bug is investigated. Check the W&B release notes for known issues with your version.

Metrics logged before wandb.init()

Metrics logged before wandb.init() is called are silently discarded. Ensure wandb.init() is called before any wandb.log() calls:

import wandb

wandb.init(project="my-project")  # must come first

for step in range(100):
    wandb.log({"loss": loss, "accuracy": acc})

Metrics logged after wandb.finish()

Any wandb.log() calls after wandb.finish() are also discarded. If you call wandb.finish() mid-script and then log more data, those metrics will not appear.

Offline mode without syncing

If WANDB_MODE=offline is set, metrics are saved locally but not uploaded until you run wandb sync. Check whether the run shows data locally in your wandb/ directory and sync it:

wandb sync wandb/run-<timestamp>-<run-id>

Non-deterministic drops in distributed training

In distributed training setups, only the process that called wandb.init() should call wandb.log(). If multiple processes log to the same run without coordination, metrics can overwrite each other or drop. Use rank checks to ensure only one process logs:

import os

if int(os.environ.get("RANK", 0)) == 0:
    wandb.log({"loss": loss})

---

Logs Metrics Experiments