Skip to content

Restructure the docs for a clearer learning path and correct stale content - #311

Open
JyotinderSingh wants to merge 2 commits into
keras-team:mainfrom
JyotinderSingh:docs-flow-learning-experience
Open

Restructure the docs for a clearer learning path and correct stale content#311
JyotinderSingh wants to merge 2 commits into
keras-team:mainfrom
JyotinderSingh:docs-flow-learning-experience

Conversation

@JyotinderSingh

Copy link
Copy Markdown
Collaborator

Description

This PR reorganizes the documentation site so that a new user understands how Kinetic works in the first two pages. It also corrects the pages against the code and rewrites the prose to be more human-like and easier to follow.

The new flow

The sidebar now has six sections, in learning order:

  1. Start Here: Getting Started, a new How Kinetic Works page, Troubleshooting, and the FAQ. Getting Started ends with a first job. How Kinetic Works defines the vocabulary: profile, cluster, node pool, job, package root, output directory.
  2. Run Jobs: data, outputs and checkpoints, dependencies, environment variables, detached jobs, batched jobs, debugging, distributed training, and profiling.
  3. Manage Infrastructure: profiles, clusters and node pools, cost, and reservations.
  4. Advanced: container images and what ships to the pod.
  5. Examples & Tutorials and Reference.

Main changes

  • Profiles from kinetic init are the primary configuration on every page. KINETIC_* variables and CLI flags are one-off overrides.
  • The container image modes now live on one advanced page. execution_modes.md and advanced.md are removed.
  • Every page was checked against the code. Corrections include the job ID format, what result() and cleanup delete, the multi-host TPU slices (tpu-v6e-8 has two hosts), the fact that Kinetic does not publish prebuilt base images, a fixed output_dir for resume, Spot pools that need spot=True, and the idle cluster costs.
  • Three example docstrings that pointed at removed setup steps or wrong environment variables are updated.

…ntent

Reorganize the sidebar into Start Here, Run Jobs, Manage Infrastructure,
Advanced, Examples, and Reference. Add a "How Kinetic Works" concepts
page. Make profiles the primary configuration mechanism on every page,
with KINETIC_* variables and flags as one-off overrides. Move the
container image modes to one advanced page (containers.md) and remove
execution_modes.md and advanced.md.

Correct the pages against the code: job ID format, cleanup and TTL
behavior, multi-host slices (tpu-v6e-8 is two hosts), no published
prebuilt base images, fixed output_dir for resume, Spot pools that need
spot=True, idle cluster costs, and the debug attach window. Rewrite all
prose in ASD-STE100 style. Fix stale example docstrings.
@codecov-commenter

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@a6adb41). Learn more about missing BASE report.

Additional details and impacted files
@@           Coverage Diff           @@
##             main     #311   +/-   ##
=======================================
  Coverage        ?   36.83%           
=======================================
  Files           ?       83           
  Lines           ?    15215           
  Branches        ?        0           
=======================================
  Hits            ?     5604           
  Misses          ?     9611           
  Partials        ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@gemini-code-assist gemini-code-assist Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request performs a comprehensive update and restructuring of the Kinetic documentation and example scripts to align with the active profile configuration workflow and modern TPU naming conventions. The review feedback identifies an architectural improvement opportunity to persist reservation and spot settings as state dimensions in StackState and load_state() to prevent fragile infrastructure updates, rather than merely documenting the limitation. Additionally, a minor markdown formatting correction was suggested for the Gemma 4 fine-tuning guide to ensure proper rendering.

Comment thread docs/guides/reservations.md
Comment thread docs/examples/gemma4_finetuning.md
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants