Skip to content

Fix W&B run reuse across ART models #618

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Kovbo merged 2 commits into from
Mar 17, 2026
Merged

Fix W&B run reuse across ART models #618

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0db8ff8
Fix wandb run reuse across models
Kovbo Mar 16, 2026
7ee516e
Remove ART wandb regression test
Kovbo Mar 16, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 17 additions & 14 deletions src/art/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -470,6 +470,7 @@ def _get_wandb_run(self) -> Optional["Run"]:
id=self.name,
config=self._wandb_config or None,
resume="allow",
reinit="create_new",
settings=wandb.Settings(
x_stats_open_metrics_endpoints={
"vllm": "http://localhost:8000/metrics",
Expand All @@ -492,18 +493,18 @@ def _get_wandb_run(self) -> Optional["Run"]:

# Define training_step as the x-axis for all metrics.
# This allows out-of-order logging (e.g., async validation for previous steps).
wandb.define_metric("training_step")
wandb.define_metric("time/wall_clock_sec")
wandb.define_metric("reward/*", step_metric="training_step")
wandb.define_metric("loss/*", step_metric="training_step")
wandb.define_metric("throughput/*", step_metric="training_step")
wandb.define_metric("costs/*", step_metric="training_step")
wandb.define_metric("time/*", step_metric="training_step")
wandb.define_metric("data/*", step_metric="training_step")
wandb.define_metric("train/*", step_metric="training_step")
wandb.define_metric("val/*", step_metric="training_step")
wandb.define_metric("test/*", step_metric="training_step")
wandb.define_metric("discarded/*", step_metric="training_step")
run.define_metric("training_step")
run.define_metric("time/wall_clock_sec")
run.define_metric("reward/*", step_metric="training_step")
run.define_metric("loss/*", step_metric="training_step")
run.define_metric("throughput/*", step_metric="training_step")
run.define_metric("costs/*", step_metric="training_step")
run.define_metric("time/*", step_metric="training_step")
run.define_metric("data/*", step_metric="training_step")
run.define_metric("train/*", step_metric="training_step")
run.define_metric("val/*", step_metric="training_step")
run.define_metric("test/*", step_metric="training_step")
run.define_metric("discarded/*", step_metric="training_step")
self._sync_wandb_config(run)
return self._wandb_run

Expand Down Expand Up @@ -562,14 +563,16 @@ def _log_metrics(
run.log(prefixed)

def _define_wandb_step_metrics(self, keys: Iterable[str]) -> None:
import wandb
run = self._wandb_run
if run is None or run._is_finished:
return

for key in keys:
if not key.startswith("costs/"):
continue
if key in self._wandb_defined_metrics:
continue
wandb.define_metric(key, step_metric="training_step")
run.define_metric(key, step_metric="training_step")
self._wandb_defined_metrics.add(key)

def _route_metrics_and_collect_non_costs(
Expand Down
Loading