{
  "device": "NVIDIA GB10",
  "torch": "2.9.1+cu130",
  "attention": "sage2",
  "resolution": [
    1344,
    768
  ],
  "frames": 124,
  "steps": 1,
  "seed": 440420,
  "text_tokens": 100,
  "warmup_runs": 1,
  "cuda_profiler_capture": true,
  "argv": [
    "tools/profile_sampling_stages.py",
    "--attention",
    "sage2",
    "--steps",
    "1",
    "--warmup-runs",
    "1",
    "--uninstrumented",
    "--cuda-profiler-capture",
    "--output",
    "/output/h3-blackwell-runtime/benchmarks/gb10-post-fc2-warmed-step-capture-20260826.json"
  ],
  "environment_switches": {
    "CUDA_DEVICE_MAX_CONNECTIONS": "1",
    "CUDA_DEVICE_MAX_COPY_CONNECTIONS": "4",
    "CUDA_HOME": "/usr/local/cuda",
    "CUDA_INC_PATH": "/usr/local/cuda/include",
    "CUDA_INJECTION64_PATH": "/opt/nsys/target-linux-sbsa-armv8/libToolsInjection64.so",
    "CUDA_MANAGED_FORCE_DEVICE_ALLOC": "1",
    "CUDA_MODULE_LOADING": "EAGER",
    "CUDA_VERSION": "13.0.2",
    "H3_FUSED_ELEMENTWISE": "1",
    "H3_MODEL_PATH": "/models/minimax_h3_ref2va_pruned_nvfp4.safetensors",
    "H3_NVFP4_FC2_LT_SPLITK1": "1",
    "H3_NVFP4_MODULATE_FUSION": "1",
    "H3_NVFP4_SCALE_BACKEND": "vortex",
    "H3_NVFP4_SCALE_VERSION": "1",
    "H3_NVFP4_SWIGLU_FUSION": "1",
    "H3_SAGE_QKV_LAYOUT": "strided_nhd",
    "TORCH_COMPILE_DISABLE": "0",
    "TORCH_CUDA_ARCH_LIST": "12.1a",
    "TORCH_EXTENSIONS_DIR": "/opt/h3-blackwell-runtime/.torch_extensions"
  },
  "elapsed_seconds": 20.90812512299999,
  "stage_trace": [],
  "checksums": [
    -197605.3125,
    528.5980224609375
  ],
  "sha256": {
    "video": "b5fc9bd43ff65f189797fd2377d5a48d8d57f64fb2774fb729dc2be478c39a4b",
    "audio": "de6da2540b639a821be7d069efcbf067b209b20de3f7daa0bb9ee28e2624764c"
  },
  "fc2_dispatch_delta": {
    "attempts": 50,
    "successes": 50,
    "fallbacks": 0
  },
  "peak_allocated_bytes": 15570401280,
  "peak_reserved_bytes": 18538823680,
  "measurement_policy": "uninstrumented sampling wall time"
}