This PR introduces LocalScore, an open-source benchmarking tool that measures LLM performance on local hardware. Key features include:

• Measures prompt processing speed, generation speed, and time-to-first-token
• Supports CPU/GPU benchmarking across Windows/macOS/Linux
• Provides a unified scoring system via geometric mean of metrics
• Optional submission to public leaderboard at localscore.ai
• Multiple execution options (standalone binary, llamafile integration)

The tool collects anonymous hardware specs and performance metrics to help users compare different setups. Includes CLI interface with options for:

• GPU backend selection
• Multiple test repetitions
• Result submission control

Built as part of the Mozilla Builders program.

Usage Examples:

# Basic benchmark
./localscore -m model.gguf

# CPU-only test with extended repetitions
./localscore -m model.gguf --cpu --extended

# Automatic result submission
./localscore -m model.gguf -y

Further details can be found in the README.