PyTorch

{
  "run": {
    "kind": "BenchmarkRun",
    "spec": {
      "status": "passed",
      "timing": {
        "latencyNs": {
          "median": 90000381
        },
        "primaryStatistic": "median"
      },
      "observedAt": "2025-08-05T00:02:07.000Z",
      "measurements": [
        {
          "unit": "ns",
          "value": 90000381,
          "metric": "latency",
          "statistic": "p20"
        },
        {
          "unit": "ns",
          "value": 90000381,
          "metric": "latency",
          "statistic": "p80"
        }
      ],
      "sourceNative": {
        "source": "liger-kernel-bench",
        "metrics": {
          "ms_20": 90.00038146972656,
          "ms_50": 90.00038146972656,
          "ms_80": 90.00038146972656
        },
        "benchmark": "fused_linear_grpo_loss_token/full",
        "externalId": "fused_linear_grpo_loss_token/full/torch/nvidia-a100-sxm4-80gb/batch4-seq1024-vocab128256-bf16/2025-08-05-00-02-07"
      },
      "protocolDigest": "sha256:5b02ee9a2140fbc6fcc1c89800f6c177ba22718c8897cb532bf0ba91867c1e58",
      "workloadDigest": "sha256:afc7095eb6ca8c7f42a31baf8283aab16705f0dc6da42337ec218fb9e523fa3e",
      "environmentDigest": "sha256:90c9e2764222f592477cfccf3373519add5a1b881563956ad058c4f99e41b6b9",
      "implementationDigest": "sha256:39aefd5bbcd914f3146a525b2dd28e8598242e7015bd3eeb1a37e7217fac6484"
    },
    "metadata": {
      "name": "liger-fused-linear-grpo-loss-token-full-torch-nvidia-a100-sxm4-80gb-batch4-seq1024-vocab128256-bf16-",
      "title": "GRPO loss, token-level importance sampling · torch · full",
      "labels": {
        "liger_version": "0.6.1"
      }
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "protocol": {
    "kind": "BenchmarkProtocol",
    "spec": {
      "harness": {
        "name": "Liger-Kernel benchmark scripts",
        "repository": "https://github.com/linkedin/Liger-Kernel"
      },
      "measurement": {
        "timer": "cuda_events",
        "primaryStatistic": "median"
      },
      "comparability": {
        "notes": "Timed: the forward pass plus the backward pass, via triton.testing.do_bench with quantiles 0.5/0.2/0.8 (median with p20/p80 spread). The upstream CSV records no CUDA, driver, or torch version; the Liger release rides each run's labels. Comparable only within one kernel, workload, timed pass, and GPU.",
        "family": "liger_kernel_bench"
      }
    },
    "metadata": {
      "name": "liger-bench-do-bench-full",
      "title": "Liger benchmark harness · full pass"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "environment": {
    "kind": "ExecutionEnvironment",
    "spec": {
      "hardware": {
        "vendor": "nvidia",
        "product": "NVIDIA A100",
        "formFactor": "SXM4-80GB",
        "architecture": "sm_80"
      },
      "software": {}
    },
    "metadata": {
      "name": "liger-bench-nvidia-a100-sxm4-80gb",
      "title": "Liger benchmark host · A100-SXM4-80GB"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  }
}