{
  "schema_version": 1,
  "recorded_at": "2026-08-30",
  "model_id": "glm-5.3-flash-nvfp4",
  "checkpoint": "LibertAIDAI/GLM-5.3-Flash-NVFP4",
  "runtime": "FreeToken",
  "vast": {
    "instance_id": 49263854,
    "machine_id": 56838,
    "host_id": 92578,
    "country": "IT",
    "hourly_price_usd_including_storage": 0.7037777778
  },
  "hardware": {
    "gpu": "NVIDIA GeForce RTX 5090",
    "gpu_vram_mib": 32607,
    "cpu": "2x AMD EPYC 7V12",
    "physical_cpu_cores": 128,
    "logical_cpu_threads": 256,
    "system_ram_gib_approx": 2048,
    "gpu_local_cpu_affinity": "0-63,128-191"
  },
  "settings": {
    "moe_backend": "hybrid",
    "moe_cpu_threads": 48,
    "memory_ratio": 0.95,
    "cuda_graph": true,
    "expert_fetch": "auto",
    "kv_reserve_tokens": 8192,
    "max_running_requests": 1
  },
  "bandwidth_gbps": {
    "cpu_stream": 74.0,
    "pcie_h2d": 24.4,
    "pcie_d2h": 28.28,
    "nvfp4_cpu_moe": 73.82,
    "overlapped_cpu": 16.32,
    "overlapped_pcie": 10.47
  },
  "qualification": {
    "target_decode_tokens_per_second": 12.0,
    "result": "pass",
    "decode_tokens_per_second": 16.83975282518497,
    "milliseconds_per_token": 59.383294,
    "measured_steps": 126,
    "measured_seconds": 7.482,
    "p50_milliseconds": 56.5688,
    "p99_milliseconds": 135.0601,
    "warm_ttft_milliseconds": 7134.645,
    "server_vram_gib": 29.207,
    "peak_gpu_temperature_c": 51,
    "cache_slots": 842,
    "cache_pages": 8474
  },
  "artifacts": {
    "benchmark_json_sha256": "d2025a75f318b5a20af0fd9c91b62238d1cb1567ffcb36829f3ee7736bfe5c6e",
    "benchmark_log_sha256": "fb1d75868c8e8772f7f5365203dacf53031d182667f14db19d169cfbdefd31a9",
    "gpu_csv_sha256": "490ecc893844f1de809efa4e5537f3991663ed697da1a309f03781f7c978b0d8",
    "bandwidth_log_sha256": "e9df2b6a124364f9fd72579f3dbbb5091e428596986c63ea3adca4af7d2e29b8"
  },
  "scope": "Single-request persistent-instance qualification; live Vast Serverless qualification remains pending."
}