Transformers
GroupNorm · batch = 128 · hidden = 512 · channels = 64 · channels_per_group = 4 · fp32 · run 01a0202b-2351…
●passedReported evidence
Not re-observed recently.
Primary measurement
73.7µsmedian
Rank 2 in its comparison group · source-native comparison · observed 2026-02-28
ms 20 0.07171200215816498 · ms 50 0.07372800260782242 · ms 80 0.0739263966679573
Identity
implementationTransformers
projectHugging Face Transformers
revisionunknown
workloadbatch = 128 · hidden = 512 · channels = 64 · channels_per_group = 4 · fp32
comparison keysha256:632687a0a3d057a5…
sourceLiger-Kernel benchmarks
external idgroup_norm/backward/huggingface/nvidia-b200/batch128-channels64-channels_per_group4-hidden512-fp32/2026-02-28-00-23-31
sha256:5cde9b90d5f90633d383292…
Correctness
Marked passed by the source; the correctness policy was not published.
Workload
batch128
hidden512
channels64
channels_per_group4
xfp32 [128, 64, 512]
biasfp32 [64]
weightfp32 [64]
definition comparatornot_asserted
Measurements
latency · median73.7 µs
latency · p2071.7 µs
latency · p8073.9 µs
Protocol
harnessLiger-Kernel benchmark scripts
timercuda_events
primaryStatisticmedian
comparabilityFamilyliger_kernel_bench
Environment
gpuNVIDIA B200 (sm_100)
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 manifestHide manifest
{
"run": {
"kind": "BenchmarkRun",
"spec": {
"status": "passed",
"timing": {
"latencyNs": {
"median": 73728
},
"primaryStatistic": "median"
},
"observedAt": "2026-02-28T00:23:31.000Z",
"measurements": [
{
"unit": "ns",
"value": 71712,
"metric": "latency",
"statistic": "p20"
},
{
"unit": "ns",
"value": 73926,
"metric": "latency",
"statistic": "p80"
}
],
"sourceNative": {
"source": "liger-kernel-bench",
"metrics": {
"ms_20": 0.07171200215816498,
"ms_50": 0.07372800260782242,
"ms_80": 0.0739263966679573
},
"benchmark": "group_norm/backward",
"externalId": "group_norm/backward/huggingface/nvidia-b200/batch128-channels64-channels_per_group4-hidden512-fp32/2026-02-28-00-23-31"
},
"protocolDigest": "sha256:080e6e66e4ec75a131aa0c9121c74b8e8eed64e7b6d621d567c6a28f79548be8",
"workloadDigest": "sha256:b0f0ba379fd1abb4c9a1f5706d6a5f3a7e580390ca44a1f2789419309e6bd16e",
"environmentDigest": "sha256:ad459e643ef802f7008f1fa8915785c905d05abbfdb5eb0328be214f8d9f3a8b",
"implementationDigest": "sha256:e8c7d555b6002524eb04f3e8a91c03603172925143b51877f4050be074e9e025"
},
"metadata": {
"name": "liger-group-norm-backward-huggingface-nvidia-b200-batch128-channels64-channels-per-group4-hidden512-",
"title": "GroupNorm · huggingface · backward",
"labels": {
"liger_version": "0.7.0"
}
},
"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 B200",
"architecture": "sm_100"
},
"software": {}
},
"metadata": {
"name": "liger-bench-nvidia-b200",
"title": "Liger benchmark host · B200"
},
"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 GroupNorm →JSON