Why is console output not captured for my run? - Weights & Biases Documentation

W&B captures your script’s stdout and stderr output and stores it as output.log on the run’s Files tab. By default, the full file is uploaded when the run finishes — so an “empty” or “missing” log is often a question of when it appears, not whether it was captured. Most reports fall into one of the situations below, and the single workaround that resolves the majority of them is console_multipart=True (SDK v0.22.3+), which writes console output to timestamped chunks and uploads each chunk as soon as it closes.

import wandb

with wandb.init(settings=wandb.Settings(
    console_multipart=True,
    console_chunk_max_bytes=1_000_000,   # rotate at ~1MB
    console_chunk_max_seconds=60,        # or every 60s, whichever first
)):
    ...

Both limits can be set together — whichever is hit first triggers an upload. Set both to 0 to upload only at run finish (the default behavior). Console capture is disabled Console capture can be turned off via settings or environment variable:

# Disabled via settings
wandb.init(settings=wandb.Settings(console="off"))

# Disabled via environment variable
WANDB_CONSOLE=off python my_script.py

Check whether either of these is set in your environment or launch configuration. To re-enable, remove the setting or set WANDB_CONSOLE=wrap. Distributed training (DDP / multiprocessing): only the owning process is captured The Logs tab records output only from the process that owns the active W&B run. In Lightning/DDP, calls like wandb.termlog() or print() from worker processes that don’t own the run will only appear in the local terminal, not in W&B. Initialize the run on rank 0 only and switch console capture to wrap, which is more robust than the default under DDP:

import wandb
from lightning.pytorch.loggers import WandbLogger

wandb_logger = WandbLogger(
    project="my_project",
    settings=wandb.Settings(console="wrap"),   # or set WANDB_CONSOLE=wrap
)
trainer = Trainer(logger=wandb_logger, strategy="ddp", devices=..., accelerator="gpu")

If wrap still leaves the Logs tab empty, fall back to console="redirect" — output will then appear in output.log on the Files tab even if it doesn’t render live in the Logs tab. The run is still running and you want the log now By default, output.log is only uploaded when a run finishes. While a run is active, the Logs tab streams output for viewing but no downloadable file appears on the Files tab, and upload cadence cannot be changed after the run has started. Enable console_multipart=True at wandb.init time (see the settings block above). Chunks appear in the Files tab under logs/ while the run is active and are downloadable individually. Run crashed before flush When a run is killed (OOM, SIGKILL, machine loss, etc.) without multipart enabled, no output.log is uploaded and the download button does not appear on the run page. console_multipart=True uploads chunks as they close, so logs written before the crash are preserved on the server. A local copy is also always written to wandb/run-<timestamp>-<run-id>/logs/output.log — check there for the full output even if the server-side file is missing. Resumed runs lose the previous output.log On older SDKs, calling wandb.init(resume="allow", id=...) overwrites the run’s single output.log — earlier log content is lost. Upgrade to SDK v0.20.1+ and pass console_multipart=True; each session then gets its own chunks under logs/, so the original output is preserved alongside the resumed output.

wandb.init(
    id=run_id,
    resume="allow",
    settings=wandb.Settings(console_multipart=True),
)

Long-running jobs and the Logs tab display limit The Logs tab in the UI applies a display limit for performance: for runs with 0–100,000 lines total, it shows the latest 10,000 lines; once a run exceeds 100,000 lines, the tab may show stale lines rather than the latest tail. This is a UI display constraint only — the full content is always in output.log (or the multipart chunks). To access all lines, download the file from the Files tab or via the API. See How do I download the console log file from a run?.

Logs Runs

Documentation Index