cirron plan

Generate execution plans for Cirron operations (compile, build) and save, compare, or replay them. Plans show what will be executed, estimated resources, and potential issues without running anything.

Usage

cirron plan <command> [options]

Commands

Command	Description
`compile`	Generate compilation plan for ML models
`build`	Generate build plan for ML projects
`diff`	Compare two saved plans
`save`	Save plans for future reference
`replay`	Execute a saved plan from file

Options (compile / build)

Option	Description	Default
`--arch, -a`	Target architecture (`cpu`, `cuda`, `gpu`, `mps`)	Auto-detected
`--index`	Path to index/manifest file	`null`
`--validate`	Run validation checks during planning	`false`
`--save`	Save plan to file	`false`
`--json`	Output plan in JSON format	`false`
`--verbose`	Show detailed plan information	`false`
`-i, --interactive`	Enhanced planning with save options	`false`

Plan compile / build

# Compilation plans
cirron plan compile
cirron plan compile --arch cuda
cirron plan compile --validate
cirron plan compile --save my-compile-plan.json

# Build plans (PyTorch CUDA/CPU, TensorFlow GPU/CPU, Scikit-learn CPU)
cirron plan build
cirron plan build --validate --save build-plan.json

Without --arch, plans pick the architecture the same way cirron compile and cirron build do: inference.device from the model config, then the project’s hardware block. On Apple Silicon, plan compile plans for mps, and plan build plans for cpu because the build runs in a Linux container. Validation covers required files (src/model.py, requirements.txt), architecture-specific checks (e.g. CUDA availability), framework compatibility, and dependency resolution. Validation failures are reported as warnings and do not stop the plan from being generated. If plan generation fails, the command exits non-zero (36 for plan compile, 34 for plan build). The plan reports environment setup, dependency resolution, model compilation/build, architecture-specific optimization, and artifact generation steps.

Plan diff

Compare two saved plans to identify added/removed/changed/unchanged configurations and dependencies.

cirron plan diff plan-a.json plan-b.json
cirron plan diff plan-a.json plan-b.json --save comparison.json

Example diff:

Dependencies Changed:
- torch==1.9.0
+ torch==2.0.0
+ numpy==1.21.0

Model Params:
- Total: 1.2M -> 2.1M (+75.0%)

Plan save

Option	Description	Default
`--all`	Save all plan types (compile, build)	`false`
`--name <filename>`	Custom filename	Auto-generated
`--description <desc>`	Description for the plan	-
`--tags <tags>`	Comma-separated tags	-
`--list`	List all saved plans	`false`
`--cleanup [days]`	Remove plans older than N days	`30`
`--verbose`	Detailed save output	`false`
`--json`	JSON output	`false`

cirron plan save compile --name "production-compile" --description "Production compile"
cirron plan save build --tags "gpu,optimized"
cirron plan save --all --name "project-baseline" --tags "baseline"
cirron plan save --list
cirron plan save --cleanup 30

Plans are stored in ~/.cirron/plans/ with timestamp, description, tags, command type, and framework metadata.

Plan replay

Option	Description	Default
`--plan <file>`	Plan file to replay (required)	-
`--validate`	Validate environment compatibility	`true`
`--dry-run`	Show what would be executed without running	`false`
`--verbose`	Detailed execution information	`false`
`--force`	Force replay despite compatibility warnings	`false`

cirron plan replay --plan build-plan.json
cirron plan replay --plan compile-plan.json --validate
cirron plan replay --plan compile-plan.json --dry-run
cirron plan replay --plan build-plan.json --force --verbose

Examples

# Plan, compare, then execute
cirron plan compile --arch cuda --save
cirron plan build --validate --save
cirron plan diff compile-plan.json build-plan.json
cirron compile --arch cuda
cirron build

# Compare configurations
cirron plan compile --arch cpu --save cpu-plan.json
cirron plan compile --arch cuda --save cuda-plan.json
cirron plan diff cpu-plan.json cuda-plan.json

# CI/CD: generate, diff against known good, replay
cirron plan compile --arch cpu --json > compile-plan.json
cirron plan diff compile-plan.json previous-compile-plan.json
cirron plan replay --plan production-compile.json --validate

Output Formats

Compilation Plan:
  Project: sentiment-analysis
  Framework: pytorch
  Architecture: cuda
  Generated: 1/15/2024, 10:30:00 AM

Compilation Steps:
  Environment setup
  Dependencies resolution
  Model compilation
  CUDA optimization
  Artifact generation

Generated Artifacts:
  models/model_cuda.pth
  artifacts/model_info.json

{
  "command": "compile",
  "framework": "pytorch",
  "architecture": "cuda",
  "timestamp": "2024-01-15T10:30:00.000Z",
  "steps": [
    { "name": "Environment Setup", "description": "Prepare Python environment", "estimatedTime": "30s" }
  ],
  "artifacts": ["models/model_cuda.pth", "artifacts/model_info.json"]
}

Troubleshooting

Error	Resolution
`No cirron.yaml found`	`cirron init my-project`
`Invalid plan files`	Regenerate with `cirron plan compile --save new-plan.json`

Use descriptive names and tags, save baselines before major changes, and clean up old plans regularly with --cleanup. Use --json for programmatic processing in CI/CD pipelines.

Cirron CLI

Commands

Utils

Templates

plan

cirron plan

Usage

Commands

Options (compile / build)

Plan compile / build

Plan diff

Plan save

Plan replay

Examples

Output Formats

Troubleshooting

​cirron plan

​Usage

​Commands

​Options (compile / build)

​Plan compile / build

​Plan diff

​Plan save

​Plan replay

​Examples

​Output Formats

​Troubleshooting

cirron plan

Usage

Commands

Options (compile / build)

Plan compile / build

Plan diff

Plan save

Plan replay

Examples

Output Formats

Troubleshooting