{
  "model_name": "Qwen/Qwen3-4B",
  "gpu_name": "NVIDIA H200",
  "attention_implementation": "sdpa",
  "batch_size": 3,
  "sequence_length": 2048,
  "num_steps": 10,
  "forward": {
    "stage_name": "forward",
    "duration_ms": 1615.8598741167225,
    "tokens_processed": 61440,
    "tokens_per_second": 38023.09902248482,
    "energy_joules": 873.9250000119209,
    "energy_per_token": 0.014224039713735693,
    "avg_power_watts": 541.9081076256928,
    "peak_memory_gb": 76.5540771484375,
    "avg_gpu_util_percent": 100.0
  },
  "backward": {
    "stage_name": "backward",
    "duration_ms": 3462.180594098754,
    "tokens_processed": 61440,
    "tokens_per_second": 17746.04135460864,
    "energy_joules": 1696.024000003934,
    "energy_per_token": 0.027604557291730693,
    "avg_power_watts": 472.8399628680292,
    "peak_memory_gb": 76.5540771484375,
    "avg_gpu_util_percent": 100.0
  },
  "optimizer": {
    "stage_name": "optimizer",
    "duration_ms": 551.849422918167,
    "tokens_processed": 61440,
    "tokens_per_second": 111334.71821915968,
    "energy_joules": 316.88299998641014,
    "energy_per_token": 0.005157600911237144,
    "avg_power_watts": 499.2301039455484,
    "peak_memory_gb": 76.5540771484375,
    "avg_gpu_util_percent": 100.0
  },
  "total_duration_ms": 5629.889891133644,
  "total_tokens": 61440,
  "total_tokens_per_second": 10913.179687005982,
  "total_energy_joules": 2886.832000002265,
  "total_energy_per_token": 0.04698619791670353,
  "timestamp": 1769149487.0005488
}