{
  "execution_mode": "offline analysis of fresh CPU INT8 model execution",
  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
  "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
  "quantized_linear_modules": 169,
  "comparison": {
    "predicted_tokens": 41,
    "baseline_token_weighted_nll": 6.035825868932212,
    "candidate_token_weighted_nll": 7.285529497193127,
    "nll_increase": 1.2497036282609146,
    "likelihood_regression": true
  },
  "limits": "Development likelihood only; no held-out task-quality, GPU performance or production acceptance."
}