{
  "status": "passed",
  "created_at": "2026-09-27T10:27:24.189892+00:00",
  "runtime": {
    "python": "3.11.12",
    "platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
    "gemmadecision": null,
    "granian": "2.8.3",
    "torch": "2.14.0",
    "transformers": "5.17.0",
    "vllm": null,
    "pydantic-ai-slim": "2.51.0",
    "httpx": "0.28.1",
    "gpu_inventory": [
      "NVIDIA H100 80GB HBM3, 580.95.05, 81559 MiB"
    ]
  },
  "backend": "torch",
  "command": [
    "python",
    "-m",
    "gemmadecision.cli",
    "serve",
    "--model-path",
    "MODEL_DIR",
    "--device",
    "cuda",
    "--port",
    "18700",
    "--backend",
    "torch",
    "--cache-size",
    "0",
    "--batch-wait-ms",
    "1",
    "--max-batch-size",
    "32"
  ],
  "model_source": "MODEL_DIR",
  "cache_size": 0,
  "batch_wait_ms": 1,
  "max_batch_size": 32,
  "max_batch_tokens": 8192,
  "accuracy_evaluated": false,
  "authentication": "random ephemeral bearer key, never written to report",
  "measurements": [
    {
      "concurrency": 1,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5416,
      "cached_pairs": 0,
      "elapsed_s": 1.1856150050000025,
      "requests_per_s": 26.990211717166932,
      "candidate_pairs_per_s": 107.96084686866773,
      "request_latency_ms_p50": 25.625442999999137,
      "request_latency_ms_p95": 70.17135694999901,
      "request_latency_ms_min": 24.37622599999756,
      "request_latency_ms_max": 179.01753900000017,
      "model_batcher_calls": 32,
      "requests_per_batcher_call": 1.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    },
    {
      "concurrency": 8,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5416,
      "cached_pairs": 0,
      "elapsed_s": 0.3418686080000022,
      "requests_per_s": 93.60321261202138,
      "candidate_pairs_per_s": 374.4128504480855,
      "request_latency_ms_p50": 81.29092599999765,
      "request_latency_ms_p95": 121.52853865000246,
      "request_latency_ms_min": 31.423168999999973,
      "request_latency_ms_max": 121.84021100000209,
      "model_batcher_calls": 8,
      "requests_per_batcher_call": 4.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    },
    {
      "concurrency": 32,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5544,
      "cached_pairs": 0,
      "elapsed_s": 0.37386616499999903,
      "requests_per_s": 85.59212626261615,
      "candidate_pairs_per_s": 342.3685050504646,
      "request_latency_ms_p50": 361.9161704999989,
      "request_latency_ms_p95": 366.5764232999999,
      "request_latency_ms_min": 57.75913400000121,
      "request_latency_ms_max": 367.0194379999998,
      "model_batcher_calls": 2,
      "requests_per_batcher_call": 16.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    }
  ],
  "server_log": "server-results.server.log",
  "batching_note": "model_batches counts scheduler calls, not individual encoder forward chunks",
  "health_after_startup": {
    "status": "ready",
    "serving_runtime": "granian-rust",
    "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
    "backend": "torch",
    "cache_size": 0,
    "temperature": 4.136820402388508,
    "max_state_tokens": 2048,
    "max_action_tokens": 768,
    "device": "cuda",
    "dtype": "bfloat16",
    "head_dtype": "float32",
    "strict": false,
    "max_batch_tokens": 8192,
    "max_batch_size": 32,
    "model_revision": "785d530221c990671f29976902540101bb9c7647",
    "verified": true,
    "queued_requests": 0,
    "completed_requests": 0,
    "model_batches": 0
  },
  "startup_to_ready_s": 12.749957202000001,
  "checks": {
    "all_three_question_types": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "answers": {
        "route": {
          "type": "choice",
          "choice": "cards",
          "probabilities": {
            "cards": 0.8173832922033145,
            "transfers": 0.024508498687310896,
            "cash": 0.02616473522167271,
            "other": 0.1319434738877019
          },
          "confidence": 0.8173832922033145,
          "scores": {
            "cards": -1.1179389953613281,
            "transfers": -15.62613296508789,
            "cash": -15.355615615844727,
            "other": -8.66240119934082
          }
        },
        "urgent": {
          "type": "noul",
          "noul": 0.2549496617556523,
          "probabilities": {
            "false": 0.7450503382443476,
            "true": 0.2549496617556523
          },
          "confidence": 0.7450503382443476,
          "scores": {
            "false": 4.559534549713135,
            "true": 0.12326765805482864
          }
        },
        "priority": {
          "type": "score",
          "score": 1.3292334118054616,
          "probabilities": {
            "0": 0.0029276603469410643,
            "1": 0.6649112675006562,
            "2": 0.33216107215240265
          },
          "confidence": 0.6649112675006562,
          "scores": {
            "0": -15.137070655822754,
            "1": 7.307041645050049,
            "2": 4.435949325561523
          }
        }
      },
      "usage": {
        "input_tokens": 334,
        "output_tokens": 0,
        "total_tokens": 334,
        "cached_pairs": 0
      },
      "temperature": 4.136820402388508,
      "probabilities_source": "softmax_ranking_scores_external_temperature",
      "latency_ms": 3755.6038860000012
    },
    "pydantic_ai_output": {
      "route": "cards",
      "urgent": false
    },
    "pydantic_ai_output_type_valid": true,
    "rank_endpoint_valid": true,
    "missing_key_status": 401,
    "overlong_state_status": 422,
    "accuracy_evaluated": false
  },
  "warmup_requests": 3,
  "health_after_measurement": {
    "status": "ready",
    "serving_runtime": "granian-rust",
    "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
    "backend": "torch",
    "cache_size": 0,
    "temperature": 4.136820402388508,
    "max_state_tokens": 2048,
    "max_action_tokens": 768,
    "device": "cuda",
    "dtype": "bfloat16",
    "head_dtype": "float32",
    "strict": false,
    "max_batch_tokens": 8192,
    "max_batch_size": 32,
    "model_revision": "785d530221c990671f29976902540101bb9c7647",
    "verified": true,
    "queued_requests": 0,
    "completed_requests": 102,
    "model_batches": 49
  },
  "simple_api": {
    "passed": true,
    "first_result": "billing",
    "second_result": "technical",
    "first_call_including_load_s": 9.876531795999998,
    "warm_decide_call_s": 0.02334669599999728,
    "ranking": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "ranked": [
        {
          "rank": 1,
          "candidate": "0",
          "text": "billing",
          "score": -4.178100109100342,
          "probability": 0.9804118154565832
        },
        {
          "rank": 2,
          "candidate": "1",
          "text": "technical",
          "score": -20.36566925048828,
          "probability": 0.019588184543416823
        }
      ],
      "usage": {
        "input_tokens": 32,
        "output_tokens": 0,
        "total_tokens": 32,
        "cached_pairs": 0
      },
      "temperature": 4.136820402388508,
      "probabilities_source": "softmax_ranking_scores_external_temperature",
      "latency_ms": 198.1146700000025
    },
    "engine": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "backend": "torch",
      "cache_size": 0,
      "temperature": 4.136820402388508,
      "max_state_tokens": 2048,
      "max_action_tokens": 768,
      "device": "cuda",
      "dtype": "bfloat16",
      "head_dtype": "float32",
      "strict": false,
      "max_batch_tokens": 8192,
      "max_batch_size": 32,
      "model_revision": "785d530221c990671f29976902540101bb9c7647",
      "verified": true
    },
    "native_pydantic_ai_local": {
      "passed": true,
      "output": "billing",
      "warm_agent_call_s": 0.1541170980000004,
      "provider_url": "local://gemmadecision",
      "provider_details": {
        "confidence": {
          "response": 0.9885687139371604
        },
        "probabilities": {
          "response": {
            "billing": 0.9885687139371604,
            "technical": 0.01143128606283964
          }
        },
        "scores": {},
        "probabilities_source": "softmax_scores",
        "confidence_note": "Ranking probabilities; validate calibration on your own data."
      },
      "shared_engine_identity_unchanged": true
    },
    "accuracy_evaluated": false
  },
  "total_elapsed_s": 35.509576861
}