{
  "model": "Qwen/Qwen2.5-0.5B-Instruct-GGUF",
  "model_revision": "9217f5db79a29953eb74d5343926648285ec7e67",
  "model_file": "qwen2.5-0.5b-instruct-q4_k_m.gguf",
  "model_sha256": "74a4da8c9fdbcd15bd1f6d01d621410d31c6fc00986f5eb687824e7b93d7a9db",
  "model_license": "Apache-2.0; upstream LICENSE retained privately with downloaded weights",
  "model_source": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/tree/9217f5db79a29953eb74d5343926648285ec7e67",
  "server": "llama.cpp b12081 / 0.8.0-dev / 8fc037672",
  "server_archive_sha256": "2a773dcda8b301e3650e5832d3e7b014afcdffdb6364d4d4e5cf3a12bcbc2eeb",
  "server_archive_digest_verified": true,
  "server_license": "MIT; upstream notice retained privately with downloaded binary",
  "server_source": "https://github.com/ggml-org/llama.cpp/releases/tag/b12081",
  "context_tokens": 1024,
  "threads": 1,
  "parallel_requests": 1,
  "batch_size": 32,
  "gpu_layers": 0,
  "costs": "No paid model API endpoint used. Existing host/subscription operating allocations are unchanged; incremental infrastructure cost is not measured. Token usage is local provider-reported, not a bill or an agent-token count.",
  "observed_at": "2026-11-03T23:57:32.238465+00:00",
  "source": "Three actual local runner checkpoints plus transient systemd unit observations; not independent evaluation or controller telemetry.",
  "completed_requests": 24,
  "provider_reported_tokens": {
    "prompt_tokens": 3396,
    "completion_tokens": 1421,
    "total_tokens": 4817
  },
  "summed_request_duration_ms": 162760,
  "finish_reasons": {
    "length": 1,
    "stop": 23
  },
  "generation_settings": [
    {
      "endpoint": "http://127.0.0.1:8094/v1/chat/completions",
      "model": "qwen2.5-0.5b-instruct-q4_k_m",
      "maxTokens": 160,
      "temperature": 0.7,
      "seed": 101,
      "timeoutSeconds": 120
    },
    {
      "endpoint": "http://127.0.0.1:8094/v1/chat/completions",
      "model": "qwen2.5-0.5b-instruct-q4_k_m",
      "maxTokens": 160,
      "temperature": 0.7,
      "seed": 202,
      "timeoutSeconds": 120
    },
    {
      "endpoint": "http://127.0.0.1:8094/v1/chat/completions",
      "model": "qwen2.5-0.5b-instruct-q4_k_m",
      "maxTokens": 160,
      "temperature": 0.7,
      "seed": 303,
      "timeoutSeconds": 120
    }
  ],
  "unit_before_stop": {
    "ActiveState": "active",
    "Result": "success",
    "MemoryPeak": "576966656",
    "CPUUsageNSec": "98156395000"
  },
  "service_scope": "Transient loopback-only model server with 60% CPU, MemoryHigh550MiB/Max600MiB, no swap, 15-minute backstop. Stopped after three rounds. MemoryPeak includes cache, not RSS. No public inference API.",
  "design": "Four newly authored fictional cases fixed before this exercise, unchanged original A/B prompt templates, three sequential rounds at temperature0.7 with requested seeds101/202/303, eight calls per round. Seeds varied intentionally, all other requested generation settings fixed. Each request has a fresh user message; no expected rules or previous responses sent. No retries or selection. New cases are not blinded or independently reviewed held-out evaluation.",
  "checkpoint_sha256": {
    "101": "d20a921b2fd805e11bb1aee53382d88d51f2e7e849943f2dd638ffb2f94af028",
    "202": "62f5b4eda8827e586b16ba5bfb394b66d80c948475c75313ae1a3426c94ccf65",
    "303": "36ed64dd92844aed6ee0533bc1299c3ab53d80033690b65d93b39f6f182ebba1"
  }
}