tp replicas task grid_points denoiser_evals expected recorded completed failed machine_qps latency_mean_s latency_p95_s machine_wall_s 2 4 fl2va 9 8 32 32 32 0 0.02428042 164.378592 384.150194 1317.934 2 4 ref2va 9 8 32 32 32 0 0.01413972 282.646937 546.260662 2263.128