lib - Module

Joule Standard Library - Machine Learning Energy-aware machine learning primitives The Joule ML module treats energy consumption as a first-class optimization target. Every operation is annotated with energy costs, and the framework provides tools for: <ul> <li>Real-time power monitoring and energy tracking</li> <li>Energy-budget constrained training</li> <li>Cost models for operation energy estimation</li> <li>Roofline analysis for identifying bottlenecks</li> <li>Energy-aware optimizers that adapt based on efficiency</li> <li>Device abstraction with thermal awareness</li> </ul> Core philosophy: “Joules per compute” matters as much as accuracy.

`fn cuda_available() -> bool`

Check if CUDA is available

Source: lib.joule:341

`fn metal_available() -> bool`

Check if Metal is available

Source: lib.joule:349

`fn neural_engine_available() -> bool`

Check if Neural Engine is available

Source: lib.joule:357

`fn best_device() -> DeviceId`

Get best available device for ML workloads

Source: lib.joule:365

`fn init() -> Result<(), device::DeviceError>`

Initialize ML subsystem

Source: lib.joule:385

`fn start_power_monitoring(interval_ms: u64)`

Start power monitoring on all devices

Source: lib.joule:391

`fn stop_power_monitoring()`

Stop power monitoring

Source: lib.joule:398

`fn total_energy() -> Joules`

Get total energy consumed across all devices

Source: lib.joule:403

`fn total_power() -> Watts`

Get current total power draw

Source: lib.joule:408

`fn parse_model_source(source: &str) -> Result<ModelSource, InferenceError>`

Parse model source string into ModelSource enum

Source: lib.joule:435

`fn estimate_model_efficiency(model_path: &str) -> Result<f64, FormatError>`

Source: lib.joule:501