{
  "schema_version": "1",
  "model": {
    "id": "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit",
    "revision": "d8a069bfa8ae87d3d468412e1034acae19b5892b",
    "architecture": "qwen3_next",
    "quantization": {
      "group_size": 64,
      "bits": 4,
      "mode": "affine"
    },
    "n_experts": 512,
    "top_k": 10,
    "moe_layers": 48
  },
  "strata": {
    "commit": "6562877",
    "mlx_lm": "0.31.3"
  },
  "mode": "streamed",
  "measured_on": {
    "chip": "Apple M5 Max",
    "ram_gb": 137.4,
    "os": "macOS 26.5.1",
    "mlx_lm": "0.31.3"
  },
  "workload": {
    "prompt_set": "bench/mixed-v1",
    "prompt": "Explain how a CPU pipeline works.",
    "generated_tokens": 140,
    "warmup_tokens": 40,
    "decoding": "greedy"
  },
  "memory": {
    "target_ram_gb": 16.0,
    "reserve_gb": 4.0,
    "available_gb": 12.0,
    "enforced_memory_gb": null,
    "full_footprint_gb": 44.84,
    "measured_peak_gb": 11.78,
    "peak_rss_gb": 2.21,
    "swap_growth_gb": 0.0,
    "compression_events": 0,
    "expert_budget": 92,
    "working_set_gb": 9.17,
    "mode": "paged"
  },
  "throughput": {
    "tokens_per_s": 11.68,
    "p50_ms": 83.0,
    "p95_ms": 125.0,
    "ssd_reads_mb_per_tok": 153.4
  },
  "fidelity": {
    "method": "decode-mode teacher-forced, paged vs resident",
    "prompt_set": "arch_fidelity/mixed-v1",
    "budget": 10,
    "positions": 106,
    "agreement_pct": 98.113,
    "perplexity_delta_pct": 0.0434,
    "high_confidence_flips": 0,
    "max_abs_dlogit": 1.875,
    "mean_kl": 0.0005621814052574337,
    "bit_exact": false
  },
  "honesty": {
    "device_real": false,
    "physical_constraint_validated": false,
    "measurement_type": "development-host benchmark",
    "memory_peak_scope": "load-through-generation",
    "throughput_is_optimistic": true
  },
  "sample_output": "A **CPU pipeline** is a technique used in modern processors to improve instruction throughput by overlapping the execution of multiple instructions. Instead of completing one instruction fully before starting the next (like in a simple non-pipelined processor), the CPU breaks down instruction execution into several stages and processes different instructions at different stages simultaneously \u2014 much like an assembly line in a factory.\n\n---\n\n### **Why Use a Pipeline?**\nWithout pipelining, the CPU would spend time waiting for one instruction to finish before starting the next. This leads to unde"
}