Wandbify skills by ngrayluna · Pull Request #647 · wandb/examples

ngrayluna · 2026-08-26T23:35:19Z

Summary

Adds tooling (Python scripts) for converting Colab/Jupyter notebooks to marimo and consolidates marimo conversion guidance into a focused W&B-specific skill.

How this works

convert -> inspect diagnostics -> clean up with an agent -> verify

scripts/convert-colab-to-marimo.py converts individual .ipynb notebooks or a batch list to marimo .py files. It runs marimo convert and an initial marimo check. Records and saves diagnostics as JSON and/or .txt files.

Outputs are written under marimo/convert/:

marimo/convert/
  <example-name>/
    <example_name>.py
    .logs/
      result.json
      marimo-*.log
  convert-summary.txt

Converted notebooks may still need cleanup before they can run successfully on molab. An agent uses the generated output files (.json) to identify blockers, then use the marimo notebook skill to fix reactive-graph issues (i.e. marimo idioms) and preserve tutorial intent (i.e. preserve teaching qualities of the notebook).

Skill updates

Replaces the previous general marimo/Jupyter conversion guidance with one repo-specific skill:

SKILL.md
references/
  convert-cleanup.md
  marimo-idioms.md
  tutorial-notebook-objectives.md
  wandb-patterns.md

Key changes:

SKILL.md: defines the conversion and cleanup workflow.
convert-cleanup.md: covers conversion diagnostics and cleanup.
marimo-idioms.md: contains marimo mechanics and reactive-notebook conventions.
tutorial-notebook-objectives.md: covers tutorial narrative, purpose
wandb-patterns.md: contains W&B SDK conventions
scripts/convert-colab-to-marimo.py handles mechanical conversion and initial validation; agents rerun marimo check during cleanup and final verification.

The previous marimo skills were too broad, generic for wandb/examples. It also removed the teaching quality with strict rules about helper functions.

Scripts added

scripts/colab_sources/make_list.py: converts the source CSV into a deduplicated notebook path list.
scripts/colab_sources/notebook_paths.txt: batch input for conversion.
scripts/convert-colab-to-marimo.py: converts notebooks, runs the initial check, and records diagnostics.

Updated notebooks

Name	Colab	Molab
Credit Scorecards with XGBoost and W&B
Using W&B Sweeps with XGBoost
W&B Tables Quickstart
lcm-diffusers
Custom Progress Callback
Huggingface wandb
Optimize Hugging Face models with Weights & Biases
Intro to Weights & Biases
Intro to Weights & Biases keras
Report API Quickstart
Use WandbEvalCallback in your Keras workflow
Use WandbMetricLogger in your Keras workflow
Use WandbModelCheckpoint in your Keras workflow
OpenAI API Autologger Quickstart
Optimize Pytorch Lightning models with Weights & Biases
Organizing Hyperparameter Sweeps in PyTorch with W&B
Simple PyTorch Integration
AlphaFold with W&B Align, Fold, Log
Hyperparameter Optimization in TensorFlow using W&B Sweeps
torchtune and wandb
Pipeline Versioning with W&B Artifacts
WandB Artifacts Time to live TTL Walkthrough
Configs in W&B
Log (Almost) Anything with W&B Media
Log a Confusion Matrix with W&B
Plot ROC Curves with W&B
zoo wandb
Train and Debug YOLOv5 Models with Weights & Biases

Copilot

Copilot was unable to review this pull request because the user who requested the review is ineligible. To be eligible to request a review, you need a paid Copilot license, or your organization must enable Copilot code review.

+@app.cell(hide_code=True)
+def _(wandb_login_form):
+    # Recreate the form after authentication is submitted, requiring a fresh training submission.
+    _connection_is_submitted = wandb_login_form.value is not None


Copilot

Copilot was unable to review this pull request because the user who requested the review is ineligible. To be eligible to request a review, you need a paid Copilot license, or your organization must enable Copilot code review.

+    # remove slow mirror from list of MNIST mirrors
+    torchvision.datasets.MNIST.mirrors = [
+        mirror for mirror in torchvision.datasets.MNIST.mirrors
+        if not mirror.startswith("http://yann.lecun.com")


Copilot

🟡 Changes recommended

Unresolved critical and moderate findings affect conversion correctness, dependencies, inputs, outputs, and runtime behavior.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (10)

marimo/convert/convert-summary.txt:99

The summary records colabs/intro/run_quickstart.ipynb as passed and points to marimo/convert/run-quickstart/run_quickstart.py, but that target is not present in the change. This makes the committed batch report inconsistent with the actual outputs; regenerate the summary together with the outputs or remove the stale passed entry.
marimo/convert/huggingface-wandb/huggingface_wandb.py:9
The notebook imports and uses torch (for example in the GPU check), but it is absent from the PEP 723 dependency block. A sandbox created from this header does not explicitly guarantee that direct runtime dependency; declare torch with a compatible constraint so this notebook is reproducible.
marimo/convert/openai-api-autologger-quickstart/openai_api_autologger_quickstart.py:108
This cell overwrites any valid environment/client credential with the literal placeholder sk-foo, so the following API calls fail unless a user edits the source. Read the key from a password form or OPENAI_API_KEY without displaying or committing it, and leave existing runtime credentials intact.
marimo/convert/pipeline-versioning-with-w-b-artifacts/marimo/session/pipeline_versioning_with_w_b_artifacts.py.json:1
This committed file is a marimo session cache containing rendered cell outputs, not notebook source or diagnostics. The repository skill explicitly excludes __marimo__/ runtime-generated files; remove this cache from the change and ignore the directory so local sessions cannot add machine-specific state.
marimo/convert/torchtune-and-wandb/torchtune_and_wandb.py:2
"-" is not a valid PEP 508 requirement. A clean uvx marimo edit --sandbox run will fail dependency resolution before this converted notebook can start; replace this placeholder with the actual runtime requirements (including marimo) during cleanup.
marimo/convert/wandb-artifacts-time-to-live-ttl-walkthrough/wandb_artifacts_time_to_live_ttl_walkthrough.py:193
This hard-codes the entity wandb-smle, so readers authenticated to another account or team cannot run the walkthrough's artifact logging step. Make the entity an input (or omit it to use the authenticated default) as required for a reusable molab tutorial.
scripts/colab_sources/make_list.py:24
The CSV's Path column also contains examples/.../*.py entries, but prepare_notebook() rejects every non-.ipynb input. Because this comprehension copies every non-empty path, the checked-in batch list produces seven guaranteed prepare_failed results, as recorded in marimo/convert/convert-summary.txt. Filter to .ipynb here or maintain a separate list for Python examples.
scripts/colab_sources/notebook_paths.txt:12
This entry does not exist in the repository, so the committed batch input always produces a prepare_failed result instead of converting the documented Keras pipeline. Update the source CSV and regenerate this list, or remove the stale entry.
scripts/convert-colab-to-marimo.py:200
The command transcript serializes raw subprocess arguments, while prepare_notebook() passes absolute source and target paths. The committed logs consequently contain machine-specific paths such as /Users/noahluna/Desktop/..., so another checkout cannot replay the recorded command and the author's local path is unnecessarily exposed. Sanitize command arguments to repository-relative paths (or omit machine-local transcripts) before writing diagnostics.
scripts/convert-colab-to-marimo.py:304
This invokes marimo convert without checking or repairing the generated PEP 723 metadata. Several committed outputs import marimo but omit it from their dependency headers, so the result is not self-contained in a clean uv environment despite the notebook skill requiring complete runtime dependencies. Add/validate the marimo requirement after conversion or report incomplete metadata as a preparation failure.

Files reviewed: 73/98 changed files
Comments generated: 10
Review effort level: Lite

@@ -0,0 +1,765 @@
+# /// script
+# dependencies = ["client.git@molecule_tables @ git+https://github.com/wandb/client.git@molecule_tables"]


@@ -0,0 +1,244 @@
+# /// script
+# dependencies = ["wandb"]


+        ],
+    )
+
+    response_1 = openai.ChatCompletion.create(**chat_request_kwargs)


@@ -0,0 +1,466 @@
+# /// script
+# dependencies = ["wandb", "wandb-workspaces"]


+@app.cell
+def _():
+    # magic command not supported in marimo; please file an issue to add support
+    # %%writefile mistral_wandb_lora.yaml
+    # tokenizer:


@@ -0,0 +1,345 @@
+# /// script
+# dependencies = ["wandb"]


+docs/models/integrations/keras.mdx,36,https://colab.research.google.com/github/wandb/examples/blob/master/colabs/keras/Use_WandbMetricLogger_in_your_Keras_workflow.ipynb,Yes,/wandb/examples/blob/master/colabs/keras/Use_WandbMetricLogger_in_your_Keras_workflow.ipynb
+docs/models/integrations/keras.mdx,74,https://colab.research.google.com/github/wandb/examples/blob/master/colabs/keras/Use_WandbModelCheckpoint_in_your_Keras_workflow.ipynb,Yes,/wandb/examples/blob/master/colabs/keras/Use_WandbModelCheckpoint_in_your_Keras_workflow.ipynb
+docs/models/integrations/keras.mdx,149,https://colab.research.google.com/github/wandb/examples/blob/master/colabs/keras/Use_WandbEvalCallback_in_your_Keras_workflow.ipynb,Yes,/wandb/examples/blob/master/colabs/keras/Use_WandbEvalCallback_in_your_Keras_workflow.ipynb
+docs/models/integrations/keras.mdx,260,https://colab.research.google.com/github/wandb/examples/blob/master/colabs/keras/Keras_pipeline_with_Weights_and_Biases.ipynb,Yes,/wandb/examples/blob/master/colabs/keras/Keras_pipeline_with_Weights_and_Biases.ipynb


+Use [`../../scripts/convert-colab-to-marimo.py`](../../scripts/convert-colab-to-marimo.py)
+to create the initial marimo notebook from a Jupyter `.ipynb`, capture
+`marimo check` output, and write a temporary `.conversion/` report directory
+for the polishing pass. The report directory includes a Markdown handoff,
+raw convert/check output, structured JSON metadata, and an event log.


+# /// script
+# requires-python = ">=3.10"
+# dependencies = [


+            "Run this script from the root of the wandb/examples repository "
+            "(the directory containing examples/marimo)."
+        )


Replace the legacy YOLOv5 repository workflow with the maintained Ultralytics package and add gated dataset preparation, detection, and W&B training interactions.

+                    "optimizer": "Adam"}
+
+    entity = wandb_entity.value.strip() or None
+    model = train_and_log(train_config, entity=entity)


+    report_5.blocks = [b]
+    assert b.text == ["Hello", " World!"]
+    assert report_5.blocks[0].text == ["Hello", " World!"]
+    initial_assignment_done = True


+    assert report_5.blocks[0].text == ["Something", " New"]
+
+    # Older releases could leave the previously assigned value here.
+    mutation_example_done = True


ngrayluna and others added 30 commits August 11, 2026 12:37

wip

c5313cf

remove vendor skill

7eeb511

wip

35d85f4

prepare-marimo-ex.sh updates

336fdc4

added table for logic

50435ea

wip

4b49b5f

clean up

86442bf

remove CSV file

0b38cef

rename

a4b689d

change script name

ed3786e

update docstrings

32cd72e

correct docstring

2f90e39

update .md

80be87a

renamed files

eb7d639

add convert txt file

e79afce

change naming

515ece0

add rough convert-cleanup.md

bba53f8

update SKILL.md so it checks logs first

755acf9

wip

e77e895

organize skills

b9642ac

add tutorial skill

8e371c2

more editing

1dd3fd8

update wb and marimo references

0457daf

add reference links as backup

29935fb

remove duplicate rule

a13a53b

wip

f07ad53

wip

d0e683b

not confuse app vs function decorator cells

3198575

add converted marimo scripts

4e60132

Run marimo-wandb-notebooks skill on file

b98df1f

Copilot AI reviewed Sep 14, 2026

github-code-quality Bot found potential problems Sep 14, 2026

View reviewed changes

update notebooks

b727ae6

Copilot AI review requested due to automatic review settings September 14, 2026 20:21

Copilot AI reviewed Sep 14, 2026

github-advanced-security Bot found potential problems Sep 14, 2026

View reviewed changes

Comment thread marimo/convert/simple-pytorch-integration/simple_pytorch_integration.py Outdated

# remove slow mirror from list of MNIST mirrors

torchvision.datasets.MNIST.mirrors = [

mirror for mirror in torchvision.datasets.MNIST.mirrors

if not mirror.startswith("http://yann.lecun.com")

Update custom_progress_callback notebook

82a09fe

Copilot AI review requested due to automatic review settings September 14, 2026 21:25

Copilot started reviewing on behalf of ngrayluna September 14, 2026 21:25 View session

Copilot AI reviewed Sep 14, 2026

View reviewed changes

Update simple_pytorch_integration notebook

ddb3f74

Copilot AI review requested due to automatic review settings September 14, 2026 21:55

Copilot started reviewing on behalf of ngrayluna September 14, 2026 21:56 View session

ktaletsk and others added 15 commits September 14, 2026 15:09

fix a small pytest x marimo naming bug

9cc8d55

Update optimize_pytorch_lightning_models_with_weights_biases

95df23e

update notebooks

b218279

docs: export marimo notebook(s) to Markdown [skip ci]

d1dd578

Clean up colab links, HTML shortcodes

14854c8

clean up report_api_quickstart

e39d2c0

cleanup artifacts and logs

976fd3c

Update report_api_quickstart.py

6b26825

Update wandb_artifacts_time_to_live_ttl_walkthrough.py

00036cf

Update hyperparameter_optimization_in_tensorflow_using_w_b_sweeps.py

2d2dac9

Update optimize_hugging_face_models_with_weights_biases.py

2f10feb

Update train_and_debug_yolov5_models_with_weights_biases.py

090ac97

Migrate YOLOv5 Tutorial to Ultralytics YOLO

0d20795

Replace the legacy YOLOv5 repository workflow with the maintained Ultralytics package and add gated dataset preparation, detection, and W&B training interactions.

fix few things in YOLO notebook

401a2b1

Update credit_scorecards_with_xgboost_and_w_b.py

074f29f

github-code-quality Bot found potential problems Sep 16, 2026

View reviewed changes

extract images bloating the notebook

b795411

Navigation Menu

Uh oh!

Wandbify skills - #647

Wandbify skills#647
ngrayluna wants to merge 54 commits into
mainfrom
wandbify_skills

ngrayluna commented Aug 26, 2026 •

edited by ktaletsk

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

		@@ -0,0 +1,765 @@
		# /// script
		# dependencies = ["client.git@molecule_tables @ git+https://github.com/wandb/client.git@molecule_tables"]

		@@ -0,0 +1,466 @@
		# /// script
		# dependencies = ["wandb", "wandb-workspaces"]

Uh oh!

Conversation

ngrayluna commented Aug 26, 2026 • edited by ktaletsk Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

How this works

Skill updates

Scripts added

Updated notebooks

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (10)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ngrayluna commented Aug 26, 2026 •

edited by ktaletsk

Loading