Profile ExecuTorch model execution with per-operator timings and ETDump export.
npx skills add https://github.com/Msabih/executorch --skill profile-msabih
Please help me install this Agent Skill. Skill: profile Source: https://github.com/Msabih/executorch/tree/main/.claude/skills/profile Command: npx skills add https://github.com/Msabih/executorch --skill profile-msabih
Profiling ExecuTorch model execution to quantify on-device performance, identify bottlenecks, and guide optimizations.
To profile a model run, enable ETDump, execute the forward pass, and inspect results with Inspector.
High-intent search queries and answers about installing and using this skill.
To profile ExecuTorch model execution timing on-device, enable ETDump, execute the forward pass, and inspect the captured results with Inspector to measure per-operator timings and latency.
ETDump is used to capture and save on-device execution traces during model profiling. It enables post-hoc analysis by allowing the Inspector tool to interpret the recorded data and pinpoint slow operators.
Yes, you can measure per-operator timings to identify inference bottlenecks. Profiling quantifies on-device performance by aggregating results across inputs to determine which layer contributes most to inference time.
Yes, the Inspector tool is required to analyze ETDump data. While ETDump captures the execution traces on-device, Inspector is used for the post-hoc analysis needed to interpret the profiling results and guide optimizations.
You can compare two model runs by profiling each execution with ETDump enabled. Aggregating the per-operator timings across inputs allows you to determine which specific layer or operation contributes most to the inference time.
Profiling model performance requires on-device runtime access to execute the forward pass. This direct runtime interaction is necessary to capture accurate timing data and generate the ETDump traces for analysis.
huahai2022Install dependencies and configure the conda environment for ExecuTorch development.
borgheiAnalyze marketing performance data with Python analytics pipelines.
huahai2022Profile ExecuTorch model execution to collect latency and operator timing metrics.
pytorchProfile ExecuTorch model execution with etdump and Inspector for operator timing.
