benchmarks/recognizer_features_int8.md

# Benchmark — `model.mlpackage`

- input `regions` shape = [128, 128, 8, 32]

| compute unit | mean | median | p95 | p99 | min | max | stdev |
|---|---:|---:|---:|---:|---:|---:|---:|
| `ALL` | 4.87 ms | 4.73 ms | 5.85 ms | 6.02 ms | 4.24 ms | 6.06 ms | 0.56 ms |
| `CPU_AND_GPU` | 4.60 ms | 4.53 ms | 5.01 ms | 5.11 ms | 4.26 ms | 5.13 ms | 0.26 ms |
| `CPU_AND_NE` | 11.02 ms | 11.04 ms | 11.18 ms | 11.21 ms | 10.85 ms | 11.21 ms | 0.11 ms |
| `CPU_ONLY` | 47.68 ms | 47.58 ms | 48.85 ms | 49.02 ms | 46.46 ms | 49.06 ms | 0.75 ms |

**Fastest:** `CPU_AND_GPU` at 4.53 ms median.