Show HN: Open dataset of real-world LLM performance on Apple Silicon (devpadapp.com)

A new open dataset has been launched, enabling users to benchmark local large language models (LLMs) on Apple Silicon while providing real-time hardware telemetry. This innovative tool addresses the fragmentation in the macOS LLM ecosystem, where existing solutions either focus solely on conversations or rely on command-line interfaces without correlating performance data with hardware metrics. The real-time dashboard features eight metric cards and seven live charts, allowing users to visualize GPU, CPU, and memory performance, track watts-per-token efficiency, and create customizable prompt presets. Significantly, this tool facilitates side-by-side A/B model comparisons, automatically enriches model metadata from sources like HuggingFace, and offers unified model management across various backends. By tracking detailed power consumption metrics and memory footprints, it empowers AI/ML developers to optimize their models for energy efficiency while providing session history with easy export options. This free and open-source application enhances accessibility for developers and researchers working within the Apple Silicon environment, potentially accelerating advancements in LLM performance and deployment.