style: apply /style-guide pass to models/track

[johndmulhausen]

Summary

This PR applies the /style-guide skill (Google Developer Style Guide + CoreWeave conventions) to the models/track documentation set, normalizing voice, terminology, placeholders, and procedural patterns across 18 files. The run was fully automated; edits are style-only and require human technical review before merge.

Files edited

• models/track/config.mdx
• models/track/create-an-experiment.mdx
• models/track/environment-variables.mdx
• models/track/jupyter.mdx
• models/track/limits.mdx
• models/track/log.mdx
• models/track/log/customize-logging-axes.mdx
• models/track/log/distributed-training.mdx
• models/track/log/log-models.mdx
• models/track/log/log-summary.mdx
• models/track/log/log-tables.mdx
• models/track/log/media.mdx
• models/track/log/plots.mdx
• models/track/log/working-with-csv.mdx
• models/track/project-page.mdx
• models/track/public-api-guide.mdx
• models/track/reproduce_experiments.mdx
• models/track/workspaces.mdx

Recommendations for technical review

Prerequisites

• Most how-to pages lack a Prerequisites section. Confirm whether to add prerequisites (W&B account, installed wandb, authenticated session, active project/entity, Python/Jupyter environment, required third-party packages) on config.mdx, create-an-experiment.mdx, jupyter.mdx, log-models.mdx, log-tables.mdx, plots.mdx, working-with-csv.mdx, and reproduce_experiments.mdx.
• jupyter.mdx: wandb.login() is shown without context about credentials or where to obtain an API key outside Colab.
• limits.mdx: "Performance" is never defined from a user perspective (page load vs. SDK overhead vs. training-loop latency); consider a symptoms→causes mapping in the intro.
• plots.mdx: The Bokeh example imports hello.wav without describing or linking it; note that readers must supply their own audio file.
• project-page.mdx: "Create a project" and "Add notes to a project" don't mention account, team membership, or role requirements.
• workspaces.mdx: The page is titled "View experiments results" but is entirely about workspaces; reconcile title vs. scope.

Verification steps

• Most procedures end without expected-outcome confirmation. Add brief "what you should see" sentences after key steps in config.mdx (config saved in Run Overview), create-an-experiment.mdx (artifact location in W&B App), jupyter.mdx (after %%wandb, wandb docker --jupyter), limits.mdx (after workspace reset to manual mode), log-models.mdx (use_model directory verification), log-summary.mdx (after define_metric and run.summary.update()), log-tables.mdx (table-in-workspace and get() confirmation), plots.mdx (chart appearance per type), working-with-csv.mdx (dashboard URL/click path), project-page.mdx (create/delete/star confirmations), reproduce_experiments.mdx (reproduced-run comparison), and workspaces.mdx (save/update template confirmation; clarify the "no details" case under "View your workspace template").
• jupyter.mdx: Briefly explain the difference between %%wandb (cell magic) and %wandb (line magic).
• limits.mdx: Throughput-examples tables lack interpretation; confirm the video-table last row (500 MB at 0.07/min = 35 MB/min) is intentional given the 40 MB/min target other rows aim for.

Technical accuracy

• config.mdx:
  • Line 138: parser.add_argument("-lr", "--learning_rate", type=int, default=0.001, ...) uses type=int with a 0.001 default — likely should be type=float.
  • Line 212: grafana_url is set to a Markdown-link string; confirm the W&B App's References section actually renders Markdown links in config values.
  • Verify current SDK behavior: wandb.init() "starts a background process that syncs and logs your data as a run" (line 32), flattening of nested config dictionaries via dot notation (line 54), and auto-loading of config-defaults.yaml from the script directory (line 231).
• environment-variables.mdx:
  • WANDB_DISABLE_GIT (line 48): confirm both latest commit and diff are always captured.
  • WANDB_RUN_ID (line 64): verify wandb (not the user) performs the non-word-character → dash conversion.
  • WANDB_QUIET vs. WANDB_SILENT (lines 65–66): clarify the precise behavioral difference.
  • WANDB_RESUME (line 62): verify semantics of allow and whether WANDB_RUN_ID is required only for allow or also must.
  • WANDB_DATA_DIR (line 45): confirm "staging artifacts" wording and that platformdirs.user_data_dir is still the default source.
  • Confirm documented defaults for WANDB_ARTIFACT_DIR, WANDB_DIR, WANDB_CACHE_DIR, and WANDB_CONFIG_DIR match current SDK behavior.
  • WANDB_DOCKER (line 50): verify wandb docker [IMAGE_NAME]:[TAG] --digest is still the current CLI syntax.
  • Multiple entries had ambiguous "we" rewritten with wandb as subject — confirm the SDK/library is the actor in each case.
• jupyter.mdx:
  • Confirm canonical form between wandb.Run.finish() and run.finish().
  • Verify the Overview tab still surfaces a Colab link.
  • Confirm %%wandb behavior when a cell runs multiple times (line 41).
  • Line 72: ipython.display(...) vs. the display symbol imported on line 76 — verify the inline reference matches IPython.display.display.
  • Line 80 comment "If cell outputs run.finish()" — confirm the comment accurately describes the trigger, since the code calls run.finish() directly.
  • Confirm wandb docker --jupyter still exists and behaves as described.
  • Verify https://wandb.me/jupyter-interact-colab still resolves.
  • Confirm bolded Overview and User Settings match current product UI labels.
  • Confirm whether wandb.init() examples should migrate to the context-manager pattern recommended by the W&B section of the style guide.
• limits.mdx:
  • Verify throughput recommendations in the three tables (Recommendations at scale, Scalar logging, Video logging) are still current for Multi-tenant Cloud.
  • Confirm: single-value < 1 MB, single-call < 25 MB, run config < 10 MB, 1,000 files per run.
  • Confirm RateLimit-Limit / RateLimit-Remaining are scaled 0–1,000.
  • Confirm the "wait at least 1 second between requests" guidance for public-API requests is still accurate.
  • Confirm wandb sync [RUN-FILE-PATH] accepts a file path (not a directory).
  • Confirm the GraphQL rate-limit description (per IP unauthorized, per user authorized, per project for some SDK requests).
  • Confirm https://wandb.me/log-hf-colab is live.
• log.mdx:
  • Line 12: verify the SDK itself (not a separate agent/process) performs both the local save and the sync to W&B cloud / private server.
  • Line 30: verify "You can't write to a specific history step. W&B only writes to the 'current' and 'next' step." reflects current SDK behavior.
  • Line 50: confirm wandb.me/code-save-colab is still canonical for the general Code Saving setting, and confirm the in-product label casing.
  • Line 53: confirm W&B (not the user/CI) is the actor uploading requirements.txt to the run's Files tab.
  • Line 85: wandb.run.summary["best_accuracy"] (lowercase run) vs. wandb.Run.log() (uppercase Run) elsewhere — confirm and apply consistently.
  • Line 63: consider linking wandb.Run.watch(model) to its reference.
• log/customize-logging-axes.mdx:
  • Lines 47–48: custom_step : int and metric_name : int use the int class as a value placeholder — semantically misleading. Consider realistic integers (i, i+1) or bracket-style placeholders.
  • Line 29: intro mentions logarithmic scale, but the worked example uses a squared scale (i**2).
  • Line 84: run.define_metric("train/*", step_metric="train/step") — confirm this is the documented signature and clarify that the first argument is the metric name / glob pattern.
• log/log-models.mdx:
  • Line 8: the ColabLink URL points to a branch-specific notebook (ken-add-new-model-reg-api). Verify this is stable and not a development fork.
  • Line 50: wandb.Run.use_model() vs. body uses of use_model and run.use_model() — confirm canonical form.
  • Example uses model.h5; verify HDF5 is still recommended over .keras.
  • Confirm run.use_model(...) / run.log_model(...) are canonical vs. module-level wandb.log_model(...).
  • Line 159: name = f"…" uses non-PEP 8 spacing around = for a keyword arg.
  • Verify the reference URI anchor #amazon-s3--gcs--azure-blob-storage-references still matches the target heading in track-external-files.
• log/log-summary.mdx:
  • "W&B Public API" tab code sample (lines ~99–137) has multiple likely bugs: api referenced without api = wandb.Api(); loop uses project_name but the defined variable is project; pd.DataFrame called although the import is import pandas (not as pd).
  • Confirm the opening paragraph's claim about tensor persistence (binary file + summary statistics: min, mean, variance, percentiles) reflects current behavior.
  • Clarify the antecedent of "It accepts the following values:" (line ~38) — summary argument vs. define_metric.
• log/log-tables.mdx:
  • Lines 88–115: the "Add data to resumed runs" example has a nested with wandb.init() as run: block inside another wandb.init() context — almost certainly a bug.
  • Confirm the literal artifact-name pattern run-[RUN-ID]-[TABLE-NAME]:[TAG] reflects the actual run-scoped artifact naming convention.
• log/plots.mdx:
  • The "Interactive custom charts" example uses wandb.plot_table(...) while other examples use wandb.plot.<chart>(...) — confirm both spellings are current.
  • vega_spec_name="carey/new_chart" references a user-specific preset; replace with a placeholder or note it must be replaced.
  • Confirm the "store fig = plt.figure() separately" guidance for the empty-plot error is still current.
• log/working-with-csv.mdx:
  • Experiments section step 2 opens a wandb.init() with block but the snippet is truncated mid-block (as run: with no body); confirm the partial snippet reads clearly.
  • The final full experiments script is missing import wandb and import pandas as pd — verify intent.
  • wandb.Artifact("iris_artifact", type="dataset") is unscoped; confirm docs recommend fully-qualified artifact names.
  • Verify each CSV row maps to one run, as stated after the line 116 rewrite.
• project-page.mdx:
  • Line ~232: verify the Usage panel actually links to consumer runs ("indicates and links to the run which output the artifact").
  • Lineage panel: verify "Explode toggle" terminology still matches the current UI.
  • Versions tab: verify the example artifact "zoo-wyhak4p0" still matches the current screenshot.
  • Confirm the UI offers a custom-color option beyond the preset palette ("Another preset or a custom color").
  • Verify the keyboard shortcut and "stop an active run" action-menu path are still accurate.
  • "TEXT AND CODE" — confirm the UI casing.
• reproduce_experiments.mdx:
  • Split-numbered procedure: source numerals 5–9 may render as 1–5 after a <Tabs> block resets MDX numbering — visual check before merge.
  • [COMMIT-SHA] placeholder previously hinted visually at a Git SHA via 40 characters; confirm the bracketed token is sufficient.
  • The GitHub tab's step 4 references a "Git state" field without a definition or screenshot.
• workspaces.mdx:
  • Line 168: verify the "v0.70 and later" threshold for deleting workspace templates on Dedicated Cloud / Self-Managed.
  • Line 62: confirm the "all users with access" statement for shared workspace views and whether private/team-only projects behave differently.
  • Lines 223–239: placeholders were converted to [ENTITY-NAME] / [WORKSPACE-URL]; confirm engineering expects bracket-style placeholders in copy/paste samples.

Missing content

• config.mdx: "below the root" terminology for nested config keys may be unclear; no documentation of precedence/merge behavior across run.config.update(), direct assignment, YAML, and argparse; no callout on which keys can be pinned, character limits, or plain-text vs. Markdown rendering; consider marking or relinking the TF v1 flags section as legacy.
• create-an-experiment.mdx: image at line 120 uses alt="Quickstart Sweeps Dashboard example" with a Sweeps-specific path on an Experiments page; the <Note> at lines 41–43 may not warrant a notice; confirm "Configure Experiments" / "Log Data During Experiments" link text matches actual page titles; line 85 uses "Learn more about..." instead of the standard "For more information, see..." pattern; decide whether "Run" should be a proper noun.
• environment-variables.mdx: Singularity section (line 75) lacks a concrete SINGULARITYENV_ example; "Run on AWS" (line 79) ends abruptly after the AWS batch link; the Singularity link points to the older LBL project, not apptainer.org / sylabs.io; the heading anchor change #running-on-aws → #run-on-aws is referenced by old slugs in ko/, ja/, and fr/ translations.
• jupyter.mdx: the bottom <Note> is used for cross-references and could be a "Next steps" / "Related" section; no guidance on ending a session cleanly or kernel restarts mid-run; bracket placeholders ([USERNAME], [PROJECT], etc.) lack a lead-in; code-fence language notebook is not in the CoreWeave standard set (bash, python, yaml, json, text); two long sentences (~36 and ~41 words) could be split.
• limits.mdx: a commented-out "Metric naming constraints" link is hidden in MDX comments; "Value size" mentions wandb.Media types are "handled differently" without a link; "Metric logging API rate limits" mentions paid vs. free plans without specifying values; three near-identical Multi-tenant Cloud warnings could be consolidated; "Browser considerations" calls out Chrome only; several concepts (Artifacts, offline syncing, public API, GraphQL) lack inline definitions / outbound links; the "Many metrics per run" procedural sentence could be promoted to a bulleted step.
• log.mdx: title/description vs. body scope mismatch — workflows may belong on subpages; stale commented-out blocks at lines 20–25, 27, and 33–39 reference step= and commit=False; document or remove commit=False references; add a pointer for non-consecutive / out-of-order logging since the doc only states you can't write to a specific history step; consider splitting the 36-word "Compare the best accuracy" sentence; "Best practices and tips" is a single external link to a W&B report.
• log/customize-logging-axes.mdx: the Globs section (line 76+) is a substantive topic without its own H2; line 97 contains a commented-out Colab link — either restore or remove; alt text (lines 26, 73) contains literal backticks; code comment (line 88) uses w/; inconsistent kwarg spacing (step_metric = custom_step vs. step_metric="train/step").
• log/log-models.mdx: no guidance on accepted file types or directory structures for log_model; no mention of size limits or large-model considerations; no description of how log_model interacts with remote reference URIs (s3://, etc.); no explanation of versioning behavior or auto-generated aliases on repeated calls; no coverage of use_model error behavior when the artifact doesn't exist; no cross-link to the "link a model to the registry" workflow; :latest alias syntax is used without explanation.
• log/log-summary.mdx: the list of valid summary values ("min", "max", "mean", "best", "last", "none") lacks one-line definitions, especially distinguishing "best" (requires objective) from "last" (default); the "Run Table" tab procedure (~lines 89–93) doesn't note how to add a summary column that isn't shown by default.
• log/log-tables.mdx: ArtifactEntry type is referenced in the JoinedTable args table without definition or link; "Visualize tables" is a single sentence; the 200,000-row override Note could be more specific about which sizes / UI operations cause issues.
• log/plots.mdx: wandb.plot.line_series xs/ys "must match exactly" doesn't clarify shared-x vs. per-line-x auto-detection; confusion-matrix bullet says "either probabilities or predictions but not both," but only preds is shown — consider a probs example; no cross-link to a quickstart establishing wandb.init basics.
• log/working-with-csv.mdx: no explanation of the 200,000-row artifact limit referenced in code comments; define_metric is mentioned but the example uses run.summary.update() — clarify or remove; link text [Sweeps Hyperparameter Tuning](/models/sweeps/) doesn't match the canonical "Sweeps"; commented-out {% embed %} blocks at lines 11 and 97–99 — remove or restore; inconsistent comment capitalization, an ellipsis, an exclamation, and 200000 without a thousands separator.
• project-page.mdx: the line ~329 Python sample is collapsed onto a single line — reformat across multiple lines and separate the import; placeholder URLs in the Star-a-project tabs mix <team> / <project-name> / <project_name> instead of the CoreWeave-preferred [PLACEHOLDER-NAME]; the Sweeps tab is a single sentence with only a screenshot; Project roles section doesn't enumerate roles or link to a reference; the Workspace tab section doesn't introduce the "Add panels" control referenced later; verify the "Confirm that your project is available." paragraph belongs under Create a project (currently stranded around line 367 pre-edit).
• reproduce_experiments.mdx: step 7 ("Optional: Create a Python virtual environment.") provides no command or link; step 1's "Name of the run you want to reproduce" prerequisite is never used explicitly in the procedure (only via branch name in the GitHub tab's step 4).
• workspaces.mdx: line 176 Colab URL contains a typo in the branch name (Update-wandb-workspaces-tuturial); lines 192–205 contain a large JSX-commented block duplicating the active bullet list above it — decide whether to delete; line 244 references "workspace API examples" without API reference detail; line 234 heading Copy a workspace \saved view` to another workspace` uses inline code for a concept (and changing it would break the anchor).

How to review

• Each file's changes are style edits only. Compare side-by-side and flag any that change technical meaning.
• Approve and merge to accept the edits, or close to reject them.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
wandb	🟢 Ready	View Preview	Jun 4, 2026, 4:49 PM

[github-actions]

📚 Mintlify Preview Links

🔗 View Full Preview

📝 Changed (18 total)

📄 Pages (18)

File	Preview
models/track/config.mdx	Config
models/track/create-an-experiment.mdx	Create An Experiment
models/track/environment-variables.mdx	Environment Variables
models/track/jupyter.mdx	Jupyter
models/track/limits.mdx	Limits
models/track/log.mdx	Log
models/track/log/customize-logging-axes.mdx	Customize Logging Axes
models/track/log/distributed-training.mdx	Distributed Training
models/track/log/log-models.mdx	Log Models
models/track/log/log-summary.mdx	Log Summary
... and 8 more files

---

🤖 Generated automatically when Mintlify deployment succeeds
📍 Deployment: b1e6a42 at 2026-06-05 19:39:02 UTC

[github-actions]

🔗 Link Checker Results

✅ All links are valid!

No broken links were detected.

Checked against: https://wb-21fd5541-style-guide-snippets-includes-20260604-122025.mintlify.app

[ngrayluna]

Please split this into two PRs: Sweeps and Snippets.

Also, some of the comments I've applied to integration PR apply here. Apply those changes and re-run?

[Copilot]

Pull request overview

This PR applies an automated style-guide normalization pass across the models/track documentation set, primarily updating wording, headings, and placeholders, and adding keywords metadata to improve consistency and discoverability.

Changes:

• Added keywords front matter and updated page introductions for clearer scope and navigation.
• Standardized voice, terminology, and placeholder formats across examples and procedures.
• Restructured several sections (headings/lists) for more consistent procedural flow.

Reviewed changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
models/track/config.mdx	Adds keywords and refines configuration guidance and section structure.
models/track/create-an-experiment.mdx	Clarifies the end-to-end experiment workflow; updates headings and phrasing.
models/track/environment-variables.mdx	Adds keywords and rewrites environment-variable descriptions for clarity/consistency.
models/track/jupyter.mdx	Adds keywords and refines notebook workflows and magic/inline rendering instructions.
models/track/limits.mdx	Adds keywords; reorganizes performance/rate-limit guidance and definitions.
models/track/log.mdx	Adds keywords; refreshes logging overview language and common workflow descriptions.
models/track/log/customize-logging-axes.mdx	Adds keywords; updates placeholders and explanations around custom step metrics.
models/track/log/distributed-training.mdx	Adds keywords; refines distributed logging approaches and troubleshooting wording.
models/track/log/log-models.mdx	Adds keywords; rewrites model logging/retrieval narrative and placeholder usage.
models/track/log/log-summary.mdx	Adds keywords; updates summary-metrics guidance and the Public API retrieval example.
models/track/log/log-tables.mdx	Adds keywords; restructures table workflow sections and updates examples/placeholders.
models/track/log/media.mdx	Adds keywords; updates prerequisites and improves explanations of media logging.
models/track/log/plots.mdx	Adds keywords; reworks headings and adds more contextual framing around plot types.
models/track/log/working-with-csv.mdx	Adds keywords; expands intro and refines CSV→Table/Artifact instructions and wording.
models/track/project-page.mdx	Adds keywords; refines project tab descriptions and UI instructions; updates SDK snippet.
models/track/public-api-guide.mdx	Adds keywords; clarifies import/export narrative and reorganizes some sections.
models/track/reproduce_experiments.mdx	Adds keywords; rewrites steps and standardizes placeholder formatting.
models/track/workspaces.mdx	Adds keywords and expands intro; refines saved views/templates and API examples language.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[johndmulhausen]

Thanks @ngrayluna! Status on both points:

• Split into Sweeps + Snippets: style: apply /style-guide pass to models/track #2728 was accidentally based on the wrong branch (it carried the models/sweeps + snippets/_includes commits). I've repointed it to the intended models/track content, so this PR is now models/track only. The sweeps and snippets passes already have their own PRs — style: apply /style-guide pass to models/sweeps #2722 and style: apply /style-guide pass to snippets/_includes #2736 — so the split is covered.
• Integration-PR feedback: Applied to models/track in 2a44a77 (preceding→previous, W&B-not-wandb as the actor, second person, active voice, wandb=W&B Python SDK on first mention). I left the frontmatter keywords in place for now since that thread's still open on the integrations PR — happy to strip them once we land on a global call.

Ready for another look when you have a sec.