{
  "schema_version": "1",
  "model": {
    "id": "mlx-community/DeepSeek-V2-Lite-Chat-4bit-mlx",
    "revision": "a33a8ad64477b9ca7a65b6dc1d8c06c997f87216",
    "architecture": "deepseek_v2",
    "quantization": {
      "group_size": 64,
      "bits": 4
    },
    "n_experts": 64,
    "top_k": 6,
    "moe_layers": 26
  },
  "strata": {
    "commit": "6562877",
    "mlx_lm": "0.31.3"
  },
  "mode": "resident",
  "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": 8.84,
    "measured_peak_gb": 11.45,
    "peak_rss_gb": 9.09,
    "swap_growth_gb": 0.0,
    "compression_events": 0,
    "expert_budget": 64,
    "working_set_gb": 8.84,
    "mode": "resident"
  },
  "throughput": {
    "tokens_per_s": 81.39,
    "p50_ms": 11.7,
    "p95_ms": 16.5,
    "ssd_reads_mb_per_tok": 6.8
  },
  "fidelity": {
    "method": "decode-mode teacher-forced, paged vs resident",
    "prompt_set": "arch_fidelity/mixed-v1",
    "budget": 6,
    "positions": 111,
    "agreement_pct": 100.0,
    "perplexity_delta_pct": 0.0,
    "high_confidence_flips": 0,
    "max_abs_dlogit": 0.0,
    "mean_kl": 0.0,
    "bit_exact": true
  },
  "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 method used by modern microprocessors to execute computer instructions. It allows several instructions to be executed in a sequence, in which each instruction starts only after the previous one has been completed. This method increases the efficiency of the CPU by allowing it to process multiple instructions at the same time.\n\nThe pipeline is divided into several stages, each of which performs a specific task. These stages include fetch, decode, execute, memory access, and write back.\n\n1. Fetch: The first stage of the pipeline fetches the next instruction from the memory.\n"
}