Skip to content

Focus the CLI docs on users and rewrite the compute guide - #17

Merged
EiffL merged 2 commits into
mainfrom
cli-docs-compute-guide
Oct 3, 2026
Merged

EiffL merged 2 commits into
mainfrom
cli-docs-compute-guide

Conversation

@EiffL

@EiffL EiffL commented Oct 3, 2026 •

Copy link
Copy Markdown
Member

This trims the Lightcone CLI docs to what users need, and rewrites the cluster guide as Configuring compute: how to write ~/.lightcone/compute.yaml for local and Slurm compute, with and without GPUs.

The guide documents the compute.yaml format from LightconeResearch/lightcone-cli#245, so it should merge once that PR is merged and released. The Installation page still pins lightcone-cli==0.5.0rc5, which predates it.

Configuring compute

  • Rewritten for users. It now covers:
    • offers that name their provider
    • the built-in local offer, and replacing it with your own local offer
    • allow_local: false
    • local GPUs from CUDA_VISIBLE_DEVICES
    • a first Slurm configuration with explicitly named clusters, then a Resources subsection on cpus, memory and GPUs (generic or with gpu_type)
    • fast and batch offers: what startup means, in a callout
    • advanced offer settings and connection_root
    • a field reference
    • examples on known Slurm computers, starting with NERSC Perlmutter
  • New diagram. ComputeLayers (components/compute-layers.tsx) shows lc run and lc materialize on top of lc compute, on top of the providers. It's built like the existing StackLayers. Fumadocs' only built-in diagram support is Mermaid, which can't draw this layout well and would need the mermaid package.
  • Renamed. cluster.mdx is now configuring-compute.mdx, and /lightcone-cli/cluster redirects to it from vercel.json. Every link into the old page and its sections is updated.
  • lc compute page updated to match: providers instead of connections, allow_local, local GPUs, and no schema_version in the JSON output.

Trimmed CLI section

  • Removed contributor pages: Developer corner, Architecture, Development setup, Testing, Extending, and the Engine internals section. The engine has no stable API.
  • Removed reference pages: the Glossary and the CLI reference index. The command pages now sit directly under Reference, shown in monospace through a small page-tree plugin in lib/source.ts.
  • Troubleshooting keeps only "lc: command not found", git identity, and filing a bug.
  • Links fixed: every link to a removed page or section is fixed or removed.

Small polish

  • Core concepts gets a lightbulb icon.
  • Guides headings lose their book icon in every section.

Checks

  • npm run build passes (131 static pages) and tsc --noEmit is clean.
  • Every compute.yaml example in the guide was loaded with lightcone-cli at bd25f44, the current head of #245, using lc compute resources. Excerpts were merged into a complete file first. The launches the guide describes were planned with lc compute launch --dry-run. Nothing was submitted to a real Slurm cluster, and the NERSC example says so.
  • The diagram was checked in light mode, dark mode and at phone width, with no console errors.

🤖 Generated with Claude Code

Trim the Lightcone CLI section to user-facing pages, and rewrite the
cluster guide as "Configuring compute": how to write
~/.lightcone/compute.yaml for local and Slurm compute, with and without
GPUs, following lightcone-cli#245's provider-named offers.

- Remove the contributor and maintainer pages: Developer corner,
  Architecture, Development setup, Testing, Extending, and the Engine
  internals section, since the engine has no stable API. Remove the
  Glossary and the CLI reference index too, and fix every link to them.
- Trim Troubleshooting to "lc: command not found", git identity, and
  filing a bug.
- Rename cluster.mdx to configuring-compute.mdx, redirect the old URL,
  and add a layered diagram of how lc reaches compute (ComputeLayers).
- Show the command pages directly under Reference, in monospace.
- Give Core concepts a lightbulb icon and drop the book icon from every
  Guides heading.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@vercel

vercel Bot commented Oct 3, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
docs Ready Ready Preview Oct 3, 2026 8:07pm UTC

Request Review

- Add a Resources subsection after the first Slurm configuration,
  explaining cpus, memory and accelerators and how launch flags match
  them, and fold the Slurm GPU subsection into it.
- Name clusters explicitly in launch examples (--name analysis).
- Explain startup in its own subsection, with a fast/batch example and a
  callout saying it is only a label for choosing between offers.
- Move Perlmutter into a closing "Examples on known Slurm computers"
  section.
- Remove "What a recipe needs" and "When a launch is refused", and the
  links from lc run and lc materialize that pointed at them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@EiffL
EiffL merged commit e8a3951 into main Oct 3, 2026
2 checks passed

This branch was successfully deployed

1 active deployment
Preview — edf06ee2 Deployed Oct 3, 2026 by vercel[bot]
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.

1 participant