[tune] model: ~/.mtplx/models/Youssofal--Qwen3.6-27B-MTPLX-Optimized-Speed
[tune] diagnostic telemetry active; use --no-telemetry for clean speed comparison
[tune] close heavy apps now for cleaner results before measurements start
[tune] fans may get loud during tuning and will be restored afterward
[max] fan tool: thermalforge (~/.mtplx/bin/thermalforge)
[max] baseline fans: min 7825 RPM, max 7830 RPM
[max] running `thermalforge max`...
[max] waiting 1.0s for the daemon to update target RPM...
[max] target accepted: 7826 RPM; waiting for actual fans to ramp...
[max] fans ramped: actual 7829-7845 RPM; target 7826 RPM
[tune] artifacts: <worktree>/outputs/cli/tune/tune-20260702-161817
[tune] running isolated candidates: AR, D3
[tune] AR (1/2) starting; log: <worktree>/outputs/cli/tune/tune-20260702-161817/ar.log
[tune] AR finished in 9.0s: 30.23 tok/s
[tune] AR telemetry: scope=generation | power pkg=23.3W cpu=3.5W ane=0.0W gpu=19.8W | freq P=4.04GHz M=1.01GHz GPU=1.21GHz | temp core_avg=58.7C core_max=68.2C gpu_avg=39.1C | util P=100.0% M=50.1% GPU=97.9% | fans=7830rpm | samples=4 | window=6.7s
[tune] settling 5.0s before D3
[tune] D3 (2/2) starting; log: <worktree>/outputs/cli/tune/tune-20260702-161817/d3.log
[tune] D3 finished in 4.4s: 81.73 tok/s
[tune] D3 telemetry: scope=generation | power pkg=63.9W cpu=3.4W ane=0.0W gpu=60.1W | freq P=3.99GHz M=1.01GHz GPU=1.60GHz | temp core_avg=60.1C core_max=68.0C gpu_avg=39.2C | util P=100.0% M=51.1% GPU=100.0% | fans=7821rpm | samples=2 | window=2.6s
MTPLX Tune
Results written to <worktree>/outputs/cli/tune/tune-20260702-161817

AR     30.2 tok/s   1.00x
D3     81.7 tok/s   2.70x  BEST
Speed shown is decode tok/s; prefill is tracked separately.

Best for this Mac: D3, 2.70x AR
Not saved: save disabled.

AR: telemetry=scope=generation | power pkg=23.3W cpu=3.5W ane=0.0W gpu=19.8W | freq P=4.04GHz M=1.01GHz GPU=1.21GHz | temp core_avg=58.7C core_max=68.2C gpu_avg=39.1C | util P=100.0% M=50.1% GPU=97.9% | fans=7830rpm | samples=4 | window=6.7s
D3: verify_ms=36.38 acceptance=MTP1 50/53 (94.34%) | MTP2 47/53 (88.68%) | MTP3 42/53 (79.25%)
D3: telemetry=scope=generation | power pkg=63.9W cpu=3.4W ane=0.0W gpu=60.1W | freq P=3.99GHz M=1.01GHz GPU=1.60GHz | temp core_avg=60.1C core_max=68.0C gpu_avg=39.2C | util P=100.0% M=51.1% GPU=100.0% | fans=7821rpm | samples=2 | window=2.6s
artifacts: <worktree>/outputs/cli/tune/tune-20260702-161817