Run the local development benchmark

Run the real PLLM roles on one machine and understand what the result does and does not measure.

Run the real local client, preparation, and inference roles:

pllm benchmark run \
  --model Qwen/Qwen2.5-0.5B-Instruct \
  --prompt "Explain why private inference matters." \
  --max-output-tokens 8 \
  --output benchmark.json

Review the exact benchmark run options before changing the model, workload, repetitions, or output behavior.

The command is headless by default, reports startup and execution progress on the terminal, writes a sanitized JSON report, and stops all three roles. Add --show-dashboard only when you want the interactive local view. Use --tiny only to test transport with generated weights. The report is not a canonical benchmark evidence record. A local run does not measure a multi-host or wide-area deployment, model quality, energy use, price, adversarial behavior, or production non-collusion.

Continue with research evidence requirements and metrics before comparing results.