amd-mixture-of-experts · submission 22003
AMD DeepSeek-style MoE layer · bs = 1 · seed = 1212 · seqlen = 8192 · dexpert = 2048 · dhidden = 7168 · nroutedexperts = 32 · nsharedexperts = 1 · nexpertspertoken = 4 · fp16 · run 01a006cd-f86d…
●passedReported evidence
Not re-observed recently.
Primary measurement
2.52smean of 12
Rank 81 in its comparison group · source-native comparison · observed 2025-05-13
best ns 2507802866 · mean ns 2521032506.3333335 · leaderboard score s 2.5197420810306608
Identity
implementationamd-mixture-of-experts · submission 22003
projectKernelBot user 160
revisionsubmission-22003
workloadbs = 1 · seed = 1212 · seqlen = 8192 · dexpert = 2048 · dhidden = 7168 · nroutedexperts = 32 · nsharedexperts = 1 · nexpertspertoken = 4 · fp16
comparison keysha256:7806408d16c4808c…
sourceGPU MODE KernelBot
external idsubmission/22003/benchmark/1
sha256:9d75f3d7573805eb2f2e133…
Correctness
Marked passed by the source; the correctness policy was not published.
Workload
bs1
seed1212
seqlen8192
dexpert2048
dhidden7168
nroutedexperts32
nsharedexperts1
nexpertspertoken4
inputfp16 [1, 8192, 7168]
definition comparatorkernelbot_eval
Measurements
latency · max2.54 s · n=12
latency · mean2.52 s · n=12
latency · min2.51 s · n=12
latency · std11.2 ms · n=12
Protocol
harnessreference-kernels eval.py
timerwall_clock
primaryStatisticmean
correctnessReferenceproblem reference implementation
comparabilityFamilygpumode_kernelbot
Environment
gpuAMD Instinct MI300X (gfx942)
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": {
"samples": 12,
"latencyNs": {
"mean": 2521032506,
"maximum": 2540616558,
"minimum": 2507802866
},
"primaryStatistic": "mean"
},
"observedAt": "2025-05-13T18:18:59.672Z",
"measurements": [
{
"unit": "ns",
"value": 11209071.028958427,
"metric": "latency",
"statistic": "std",
"sampleCount": 12
}
],
"sourceNative": {
"source": "gpumode-kernelbot",
"metrics": {
"best_ns": 2507802866,
"mean_ns": 2521032506.3333335,
"leaderboard_score_s": 2.5197420810306608
},
"benchmark": "amd-mixture-of-experts/1",
"externalId": "submission/22003"
},
"protocolDigest": "sha256:7c66da4b5866502eb10444a7ba013d0f6bdf1620c55847945008cad2fbf98aec",
"workloadDigest": "sha256:ac367795818a90c484028c3c89eedd0da31e56916f619af4f222867436f741ab",
"environmentDigest": "sha256:f722892449ed976f3e21b12df8b92b34573fc232883fec6a2f3da399a00abdd5",
"implementationDigest": "sha256:7d56f8d58c62b709f1e56e94d0a79d060ac6a84eaea0954e1decc650e7f371b3"
},
"metadata": {
"name": "kernelbot-22003-bs1-dexpert2048-dhidden7168-nexpertspertoken4-nroutedexperts32-nsharedexperts1-seed1",
"title": "amd-mixture-of-experts · submission 22003 · case 1",
"labels": {
"torch": "2.8.0.dev20250512+rocm6.3",
"platform": "Linux-5.15.0-136-generic-x86_64-with-glibc2.35"
}
},
"apiVersion": "kernelindex.dev/v1alpha1"
},
"protocol": {
"kind": "BenchmarkProtocol",
"spec": {
"harness": {
"name": "reference-kernels eval.py",
"repository": "https://github.com/gpu-mode/reference-kernels"
},
"correctness": {
"reference": "problem reference implementation",
"comparator": "kernelbot_eval"
},
"measurement": {
"timer": "wall_clock",
"primaryStatistic": "mean"
},
"comparability": {
"notes": "Per-case wall-clock statistics as published in the KernelBot dataset. Comparable only within one leaderboard problem and GPU type; runner software versions ride each run's labels.",
"family": "gpumode_kernelbot"
}
},
"metadata": {
"name": "gpumode-kernelbot-eval",
"title": "GPU MODE KernelBot evaluation harness"
},
"apiVersion": "kernelindex.dev/v1alpha1"
},
"environment": {
"kind": "ExecutionEnvironment",
"spec": {
"hardware": {
"vendor": "amd",
"product": "AMD Instinct MI300X",
"architecture": "gfx942"
},
"software": {}
},
"metadata": {
"name": "gpumode-kernelbot-amd-instinct-mi300x",
"title": "KernelBot runner · AMD Instinct MI300X"
},
"apiVersion": "kernelindex.dev/v1alpha1"
}
}Cite this record (permalink, digest, access date)
Report an issue with this run
Published 2026-08-15 · GPU MODE KernelBot · June 9 Researcher Reciprocity License v1.0All results for AMD DeepSeek-style MoE layer →JSON