edgepulse_cli

Standalone CLI for EdgePulse — trace on-device AI models and compare inference results from any terminal or CI/CD pipeline. No Flutter required.

Installation

dart pub global activate edgepulse_cli

Quick Start

Trace a model

edgepulse trace --model model.gguf --runs 20 --output trace.json

Compare two trace files

edgepulse compare baseline.json stressed.json

Get help

edgepulse --help
edgepulse trace --help
edgepulse compare --help

Features

  • trace: Execute N inference traces, capturing memory snapshots, thermal status, layer timing, and export to JSON, Markdown, or CSV.
  • compare: Compare baseline vs target traces with statistical summary (p50/p95/p99) and delta analysis.
  • ANSI color output and terminal formatting.
  • Exit codes suitable for CI/CD pipeline automated regression checking.

License

MIT License. See LICENSE for details.

Libraries