Run 01a0202b · KernelIndex

Primary measurement

18.7msmedian

Rank 4 in its comparison group · source-native comparison · observed 2025-11-11

Compare with #1 →

ms 20 18.729759216308594 · ms 50 18.729759216308594 · ms 80 18.729759216308594

Identity

implementationTransformers

projectHugging Face Transformers

revisionunknown

workloadseq = 8192 · batch = 2 · hidden = 2048 · intermediate = 4096 · bf16

comparison keysha256:e1742ea089aa75ea…

sourceLiger-Kernel benchmarks

external idtiled_swiglu/backward/deepspeed_tiled/nvidia-geforce-rtx-4090/batch2-hidden2048-intermediate4096-seq8192-bf16/2025-11-11-06-23-20

sha256:7fb008634e5ab23edf430da…

Correctness

Marked passed by the source; the correctness policy was not published.

Workload

seq8192

batch2

hidden2048

intermediate4096

xbf16 [2, 8192, 2048]

w_upbf16 [4096, 2048]

w_downbf16 [2048, 4096]

w_gatebf16 [4096, 2048]

definition comparatornot_asserted

Measurements

latency · median18.7 ms

latency · p2018.7 ms

latency · p8018.7 ms

Protocol

harnessLiger-Kernel benchmark scripts

timercuda_events

primaryStatisticmedian

comparabilityFamilyliger_kernel_bench

Environment

gpuNVIDIA GeForce RTX 4090 (sm_89)

Artifacts

No artifacts published with this run.

Replications and notes

No attestations yet.

Community attestations. They never change the evidence level; only a KernelIndex-controlled rerun does.

Add a reproduction or note

Canonical manifest

Show manifest

{
  "run": {
    "kind": "BenchmarkRun",
    "spec": {
      "status": "passed",
      "timing": {
        "latencyNs": {
          "median": 18729759
        },
        "primaryStatistic": "median"
      },
      "observedAt": "2025-11-11T06:23:20.000Z",
      "measurements": [
        {
          "unit": "ns",
          "value": 18729759,
          "metric": "latency",
          "statistic": "p20"
        },
        {
          "unit": "ns",
          "value": 18729759,
          "metric": "latency",
          "statistic": "p80"
        }
      ],
      "sourceNative": {
        "source": "liger-kernel-bench",
        "metrics": {
          "ms_20": 18.729759216308594,
          "ms_50": 18.729759216308594,
          "ms_80": 18.729759216308594
        },
        "benchmark": "tiled_swiglu/backward",
        "externalId": "tiled_swiglu/backward/deepspeed_tiled/nvidia-geforce-rtx-4090/batch2-hidden2048-intermediate4096-seq8192-bf16/2025-11-11-06-23-20"
      },
      "protocolDigest": "sha256:080e6e66e4ec75a131aa0c9121c74b8e8eed64e7b6d621d567c6a28f79548be8",
      "workloadDigest": "sha256:b2819cdcc499df46a50b274b08a506244f376f13e54ff4ca6738a7159e423e2f",
      "environmentDigest": "sha256:e2646efafa5e5fbd428b49a99b1c076c10d51654a549c4e50ec36da7662d4328",
      "implementationDigest": "sha256:70659a76da921ebc2777c752d1b601f56545c50a49056f176d2bf35d29374512"
    },
    "metadata": {
      "name": "liger-tiled-swiglu-backward-deepspeed-tiled-nvidia-geforce-rtx-4090-batch2-hidden2048-intermediate40",
      "title": "SwiGLU MLP (tiled benchmark) · deepspeed_tiled · backward",
      "labels": {
        "liger_version": "0.6.3"
      }
    },
    "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 backward pass of the module output given a random upstream gradient, 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-backward",
      "title": "Liger benchmark harness · backward pass"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "environment": {
    "kind": "ExecutionEnvironment",
    "spec": {
      "hardware": {
        "vendor": "nvidia",
        "product": "NVIDIA GeForce RTX 4090",
        "architecture": "sm_89"
      },
      "software": {}
    },
    "metadata": {
      "name": "liger-bench-nvidia-geforce-rtx-4090",
      "title": "Liger benchmark host · GeForce RTX 4090"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  }
}

Cite this record (permalink, digest, access date)

Report an issue with this run

Published 2026-08-20 · Liger-Kernel benchmarks · BSD-2-ClauseAll results for SwiGLU MLP →JSON