swanlab

name	swanlab
description	Query SwanLab experiment metrics and run status. Use when checking training/producer metrics, comparing runs, or investigating experiment health. Keywords: swanlab, metrics, experiment, loss, reward, training progress, wandb.

SwanLab

Query experiment metrics from SwanLab. All commands run via pixi run -e dev python3 -c "...".

Authentication

Already configured. Default workspace: junjayzhang, project: dev.

Quick Reference

List runs

import swanlab
api = swanlab.Api()

# All running experiments
for r in api.runs(path='junjayzhang/dev', filters={'state': 'RUNNING'}):
    print(f'{r.name} | {r.id} | {r.state} | {r.created_at}')

# Filter by state: RUNNING, FINISHED, CRASHED, ABORTED

Get metrics from a run

run = api.run(path='junjayzhang/dev/<experiment_id>')
df = run.metrics(keys=['<metric_key>'], sample=50)  # returns pandas DataFrame

Known metric keys

Producer runs (*_producer):

• group/reward_mean, group/reward_min, group/reward_max
• group/n_truncated, group/truncated_ratio
• group/n_non_truncated, group/success_ratio
• group/length_mean, group/length_min, group/length_max
• group/output_tokens_mean, group/output_tokens_max
• group/python_calls_mean, group/python_errors_total
• group/n_context_exhausted, group/n_max_turns

Train runs (*_train):

• loss_metrics/global_avg_loss, loss_metrics/global_max_loss
• time_metrics/end_to_end(s), time_metrics/data_loading(s), time_metrics/data_loading(%)
• optimization/learning_rate (or optimization/scheduler_0/learning_rate)

Run properties

run.name          # str
run.id            # experiment_id
run.state         # RUNNING | FINISHED | CRASHED | ABORTED
run.created_at    # ISO 8601
run.url           # web UI link

Usage Pattern

When asked to check experiment metrics or training progress, use this skill to:

1. List runs to find the relevant experiment ID
2. Fetch specific metrics with run.metrics(keys=[...], sample=N)
3. Report trends (is loss decreasing? is truncation ratio increasing? etc.)

Always wrap API calls in pixi run -e dev python3 -c "..." since swanlab is only in the dev environment.