{ "machine": "Apple M5 Pro, macOS 27.0, compute units ALL", "request": "three-head email triage request (benchmark-latency.py), 10 warmups, 100 calls, Python coremltools predict", "gliner2_decide_classification_fp16_L128_H4_K8.mlpackage": { "p50_ms": 8.18, "p95_ms": 10.33 }, "gliner2_decide_classification_w8_L128_H4_K8.mlpackage": { "p50_ms": 10.09, "p95_ms": 12.98 }, "gliner2_decide_classification_lut6_L128_H4_K8.mlpackage": { "p50_ms": 9.89, "p95_ms": 14.45 }, "gliner2_decide_classification_fp16_L256_H4_K32.mlpackage": { "p50_ms": 14.71, "p95_ms": 16.69 }, "gliner2_decide_classification_w8_L256_H4_K32.mlpackage": { "p50_ms": 20.58, "p95_ms": 38.79 }, "gliner2_decide_classification_lut6_L256_H4_K32.mlpackage": { "p50_ms": 17.58, "p95_ms": 24.61 }, "gliner2_decide_classification_fp16_L512_H4_K32.mlpackage": { "p50_ms": 44.02, "p95_ms": 98.99 }, "gliner2_decide_classification_w8_L512_H4_K32.mlpackage": { "p50_ms": 42.93, "p95_ms": 66.93 }, "gliner2_decide_classification_lut6_L512_H4_K32.mlpackage": { "p50_ms": 62.95, "p95_ms": 71.11 }, "cpu_and_ne": { "gliner2_decide_classification_fp16_L256_H4_K32.mlpackage": { "p50_ms": 689.15, "p95_ms": 1266.79, "calls": 30 } } }