cirron compile

Compile your ML models for production deployment with architecture-specific optimizations and validation.

Usage

cirron compile [options]

Options

Option	Description	Default
`--arch, -a`	Target architecture (`cpu`, `cuda`, `gpu`, `mps`)	Auto-detected
`--index`	Path to index/manifest file (JSON or YAML)	`null`
`--validate`	Run pre-compilation validation	`false`
`--strict`	Abort with exit code 35 when validation fails (otherwise validation issues are warnings)	`false`
`-i, --interactive`	Step-by-step compilation confirmations	`false`

Architecture and Framework Support

Architecture	Frameworks	Notes
`cpu`	PyTorch, TensorFlow, Scikit-learn	Lightweight, compatible with all frameworks
`cuda`	PyTorch	GPU-accelerated, validates CUDA availability
`gpu`	TensorFlow	GPU-accelerated, validates GPU availability
`mps`	PyTorch	Apple Silicon GPU (Metal), validates MPS availability

Without --arch, the architecture comes from inference.device in the model config, then from the project’s hardware block. PyTorch on Apple Silicon hardware (type: gpu, architecture: arm64, which the “Apple Silicon” preset in cirron hardware writes) resolves to mps. Compilation reads src/model.py, optimizes for the target architecture, and writes:

PyTorch → models/model_{arch}.pth
TensorFlow → models/model_{arch}/
Scikit-learn → models/model_{arch}.joblib (via joblib)

Metadata is written to artifacts/model_info.json.

Validation

With --validate, the command verifies required files (src/model.py, requirements.txt), Python version compatibility, architecture-specific requirements (e.g. CUDA availability), model instantiation, and index file format. Every failed check is reported. Without --strict the failures are warnings and compilation continues; with --strict the command stops before compiling and exits with code 35 (VALIDATION_FAILED). If compilation itself fails, the command exits non-zero (36, COMPILE_FAILED, unless a more specific code applies), with or without --strict, so CI jobs fail when no artifacts were produced.

Index Files

{
  "features": ["feature1", "feature2", "feature3"],
  "dataTypes": {
    "feature1": "float",
    "feature2": "int",
    "feature3": "string"
  }
}

YAML form is also accepted. Pass with --index config/index.json or --index config/index.yaml.

Examples

# Auto-detect architecture
cirron compile

# CPU compilation with validation
cirron compile --arch cpu --validate

# PyTorch with CUDA, strict
cirron compile --arch cuda --strict --validate

# TensorFlow with GPU
cirron compile --arch gpu

# PyTorch on an Apple Silicon Mac
cirron compile --arch mps --validate

# Compile with custom index
cirron compile --validate --index data/index.yaml

Output

models/
├── model_cpu.pth          # PyTorch CPU model
├── model_cuda.pth         # PyTorch CUDA model
├── model_mps.pth          # PyTorch Apple Silicon (MPS) model
└── model_gpu              # TensorFlow GPU model

artifacts/
├── model_info.json        # Model metadata
└── index_config.json      # Index configuration (if provided)

{
  "framework": "pytorch",
  "architecture": "cuda",
  "parameters": 1234567,
  "compilation_time": "2024-01-15T10:30:00Z"
}

Troubleshooting

Error	Resolution
`No cirron.yaml found`	Run `cirron init <name>` to initialize the project
`Required file missing: src/model.py`	Create the model file and verify it imports cleanly
`Python 3.9+ required, found 3.7`	Upgrade Python or update `pythonVersion` in `cirron.yaml`
`CUDA not available for PyTorch`	Switch to `--arch cpu`, or install CUDA and verify with `nvidia-smi`
`MPS not available for PyTorch`	Needs an Apple Silicon Mac with a PyTorch build that supports MPS; verify with `python3 -c "import torch; print(torch.backends.mps.is_available())"`, or use `--arch cpu`
`MPS architecture selected but hardware configuration is not Apple Silicon`	`mps` needs a `hardware` block with `type: gpu` and `architecture: arm64`; run `cirron hardware` and pick “Apple Silicon”, or choose another `--arch`
`MPS architecture is only supported for PyTorch`	TensorFlow uses `--arch gpu`; Scikit-learn uses `--arch cpu`
`Compilation failed`	Run `python3 src/model.py` to surface errors; check `pip install -r requirements.txt`

Integration

# GitHub Actions
- name: Compile Model
  run: |
    cirron compile --validate --arch cpu
    cirron compile --validate --arch cuda

FROM python:3.9 as compile
COPY . /app
WORKDIR /app
RUN cirron compile --arch cpu

FROM python:3.9-slim
COPY --from=compile /app/models /app/models
COPY --from=compile /app/artifacts /app/artifacts

Cirron CLI

Commands

Utils

Templates

compile

cirron compile

Usage

Options

Architecture and Framework Support

Validation

Index Files

Examples

Output

Troubleshooting

Integration

​cirron compile

​Usage

​Options

​Architecture and Framework Support

​Validation

​Index Files

​Examples

​Output

​Troubleshooting

​Integration

cirron compile

Usage

Options

Architecture and Framework Support

Validation

Index Files

Examples

Output

Troubleshooting

Integration