{
  "model_name": "Qwen/Qwen3-4B",
  "gpu_name": "NVIDIA H100",
  "attention_implementation": "sdpa",
  "batch_size": 3,
  "sequence_length": 2048,
  "num_steps": 10,
  "forward": {
    "stage_name": "forward",
    "duration_ms": 1790.2467511594296,
    "tokens_processed": 61440,
    "tokens_per_second": 34319.29143857359,
    "energy_joules": 981.029000043869,
    "energy_per_token": 0.01596726888092235,
    "avg_power_watts": 520.9058508009567,
    "peak_memory_gb": 76.45208740234375,
    "avg_gpu_util_percent": 100.0
  },
  "backward": {
    "stage_name": "backward",
    "duration_ms": 3854.5540031045675,
    "tokens_processed": 61440,
    "tokens_per_second": 15939.587290906931,
    "energy_joules": 1953.71099999547,
    "energy_per_token": 0.03179868164055127,
    "avg_power_watts": 491.5443624439596,
    "peak_memory_gb": 76.45208740234375,
    "avg_gpu_util_percent": 100.0
  },
  "optimizer": {
    "stage_name": "optimizer",
    "duration_ms": 899.9840868636966,
    "tokens_processed": 61440,
    "tokens_per_second": 68267.87372886644,
    "energy_joules": 365.9209999740124,
    "energy_per_token": 0.005955745442285358,
    "avg_power_watts": 377.8756124501158,
    "peak_memory_gb": 76.45208740234375,
    "avg_gpu_util_percent": 100.0
  },
  "total_duration_ms": 6544.784841127694,
  "total_tokens": 61440,
  "total_tokens_per_second": 9387.627170553957,
  "total_energy_joules": 3300.6610000133514,
  "total_energy_per_token": 0.053721695963758975,
  "timestamp": 1769149234.99943
}