Compare

Compare up to five generated manifests or AIPerf replay results in one executive-ready view — ISL shape, reuse, and observed latency side by side.

No artifacts loaded

Upload generated manifest.json or normalized replay_result.json files, or add a workspace run.

Trace comparison report

ISL distribution and workload shape

ISL distribution comparison

Grouped vertical bars make cross-trace bucket differences easier to scan

Shared ISL distribution

Detailed bucket-by-bucket bars from each artifact

Comparison breakdown

Generated and augmented traces are simulations for planning and replay experiments. Replay results reflect the specific server, model, context window, AIPerf version, and run options used during profiling.

Generated and augmented traces are simulations for planning and replay experiments. Replay results reflect the specific server, model, context window, AIPerf version, and run options used during profiling.