Skip to content
KernelIndex
Search⌘K
Sign in

amd-mixture-of-experts · submission 23963

KernelBot user 7 · python · License unknown
Kernel source · 1 lines ↓

Use it

Vendorable · source mirrored · license unknownView source →

No package. Vendor the mirrored source: 1 lines, June 9 Researcher Reciprocity License v1.0.

amd-fp8-moe-myself.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-mixture-of-experts-23963?include=source"
interfacepython
Compatibility
measured onAMD Instinct MI300X
declared hardwareAMD Instinct MI300X
architecturesgfx942
dtypesfp16

Benchmark evidence

2 measurements across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
AMD DeepSeek-style MoE layerfp16 · [1, 2048, 7168]
AMD Instinct MI300X
699.8ms
#80 of 118
2025-05-18
Run →
AMD DeepSeek-style MoE layerfp16 · [1, 8192, 7168]
AMD Instinct MI300X
2.77s
#82 of 118
2025-05-18
Run →

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:351ccefc44c43ba628d88a32f48a3c99891be5342bc06dc886ca2a85c3c588c4
license declaredunknown
license concludedunknown
imported2026-08-15

Kernel source

amd-fp8-moe-myself.py1 lines
aW1wb3J0IG1hdGgNCmltcG9ydCB0b3JjaA0KaW1wb3J0IHRvcmNoLm5uIGFzIG5uDQppbXBvcnQgdG9yY2gubm4uZnVuY3Rpb25hbCBhcyBGDQpmcm9tIHR5cGluZyBpbXBvcnQgRGljdCwgVHVwbGUsIE9wdGlvbmFsDQpmcm9tIHRhc2sgaW1wb3J0IGlucHV0X3QsIG91dHB1dF90DQoNCmRlZiBjdXN0b21fa2VybmVsKGRhdGE6IGlucHV0X3QpIC0+IG91dHB1dF90Og0KICAgICIiIg0KICAgIFN1Ym1pc3Npb24gdGVtcGxhdGUgZm9yIERlZXBTZWVrLXN0eWxlIE1peHR1cmUgb2YgRXhwZXJ0cyB1c2luZyBQeVRvcmNoLg0KDQogICAgQXJnczoNCiAgICAgICAgZGF0YTogVHVwbGUgb2YgKGlucHV0OiB0b3JjaC5UZW5zb3IsIHdlaWdodHM6IERpY3Rbc3RyLCB0b3JjaC5UZW5zb3JdLCBjb25maWc6IERpY3QpDQogICAgICAgICAgICAtIGlucHV0OiBJbnB1dCB0ZW5zb3Igb2Ygc2hhcGUgW2JhdGNoX3NpemUsIHNlcV9sZW4sIGhpZGRlbl9zaXplXQ0KICAgICAgICAgICAgLSB3ZWlnaHRzOiBEaWN0aW9uYXJ5IGNvbnRhaW5pbmcgbW9kZWwgd2VpZ2h0cw0KICAgICAgICAgICAgLSBjb25maWc6IERpY3Rpb25hcnkgY29udGFpbmluZyBtb2RlbCBjb25maWd1cmF0aW9uIHBhcmFtZXRlcnMNCg0KICAgIFJldHVybnM6DQogICAgICAgIFR1cGxlIGNvbnRhaW5pbmc6DQogICAgICAgICAgICAtIG91dHB1dDogUHJvY2Vzc2VkIHRlbnNvciBbYmF0Y2hfc2l6ZSwgc2VxX2xlbiwgZF9tb2RlbF0NCiAgICAgICAgICAgIC0gYXV4X2RhdGE6IERpY3Rpb25hcnkgd2l0aCBhdXhpbGlhcnkgZGF0YQ0KICAgICIiIg0KICAgIGlucHV0X3RlbnNvciwgd2VpZ2h0cywgY29uZmlnID0gZGF0YQ0KDQogICAgZF9oaWRkZW4gPSBjb25maWdbImRfaGlkZGVuIl0NCiAgICBkX2V4cGVydCA9IGNvbmZpZ1siZF9leHBlcnQiXQ0KICAgIG5fcm91dGVkX2V4cGVydHMgPSBjb25maWdbIm5fcm91dGVkX2V4cGVydHMiXQ0KICAgIG5fc2hhcmVkX2V4cGVydHMgPSBjb25maWdbIm5fc2hhcmVkX2V4cGVydHMiXQ0KICAgIG5fZXhwZXJ0c19wZXJfdG9rZW4gPSBjb25maWdbIm5fZXhwZXJ0c19wZXJfdG9rZW4iXQ0KICAgIGJhdGNoX3NpemUgPSBjb25maWdbImJhdGNoX3NpemUiXQ0KICAgIHNlcV9sZW4gPSBjb25maWdbInNlcV9sZW4iXQ0KDQogICAgcm91dGVyID0gdG9yY2guZW1wdHkoZF9oaWRkZW4sIG5fcm91dGVkX2V4cGVydHMsIGR0eXBlPXRvcmNoLmZsb2F0MTYsIGRldmljZT0nY3VkYScpDQogICAgZ2F0ZSA9IHRvcmNoLmVtcHR5KGRfaGlkZGVuLCBkX2V4cGVydCAqIChuX3JvdXRlZF9leHBlcnRzICsgbl9zaGFyZWRfZXhwZXJ0cyksIGR0eXBlPXRvcmNoLmZsb2F0MTYsIGRldmljZT0nY3VkYScpDQogICAgdXAgPSB0b3JjaC5lbXB0eShkX2hpZGRlbiwgZF9leHBlcnQgKiAobl9yb3V0ZWRfZXhwZXJ0cyArIG5fc2hhcmVkX2V4cGVydHMpLCBkdHlwZT10b3JjaC5mbG9hdDE2LCBkZXZpY2U9J2N1ZGEnKQ0KICAgIGRvd24gPSB0b3JjaC5lbXB0eShkX2V4cGVydCAqIChuX3JvdXRlZF9leHBlcnRzICsgbl9zaGFyZWRfZXhwZXJ0cyksIGRfaGlkZGVuLCBkdHlwZT10b3JjaC5mbG9hdDE2LCBkZXZpY2U9J2N1ZGEnKQ0KICAgIGRlbnNlID0gdG9yY2guZW1wdHkoYmF0Y2hfc2l6ZSwgc2VxX2xlbiwgZF9oaWRkZW4gKiAobl9yb3V0ZWRfZXhwZXJ0cyArIG5fc2hhcmVkX2V4cGVydHMpLCBkdHlwZT10b3JjaC5mbG9hdDE2LCBkZXZpY2U9J2N1ZGEnKQ0KICAgIG91dHB1dCA9IHRvcmNoLnplcm9zKGJhdGNoX3NpemUsIHNlcV9sZW4sIGRfaGlkZGVuLCBkdHlwZT10b3JjaC5mbG9hdDE2LCBkZXZpY2U9J2N1ZGEnKQ0KDQogICAgcm91dGVyID0gd2VpZ2h0c1sncm91dGVyLndlaWdodCddLnQoKQ0KICAgIGZvciBpIGluIHJhbmdlKG5fcm91dGVkX2V4cGVydHMpOg0KICAgICAgICBnYXRlWzosIGkgKiBkX2V4cGVydDooaSArIDEpICogZF9leHBlcnRdID0gd2VpZ2h0c1tmJ2V4cGVydHMue2l9LjAud2VpZ2h0J10NCiAgICAgICAgdXBbOiwgaSAqIGRfZXhwZXJ0OihpICsgMSkgKiBkX2V4cGVydF0gPSB3ZWlnaHRzW2YnZXhwZXJ0cy57aX0uMS53ZWlnaHQnXQ0KICAgICAgICBkb3duW2kgKiBkX2V4cGVydDooaSArIDEpICogZF9leHBlcnQsIDpdID0gd2VpZ2h0c1tmJ2V4cGVydHMue2l9LjIud2VpZ2h0J10NCiAgICBnYXRlWzosIG5fcm91dGVkX2V4cGVydHMgKiBkX2V4cGVydDpdID0gd2VpZ2h0c1snc2hhcmVkX2V4cGVydHMuMC53ZWlnaHQnXQ0KICAgIHVwWzosIG5fcm91dGVkX2V4cGVydHMgKiBkX2V4cGVydDpdID0gd2VpZ2h0c1snc2hhcmVkX2V4cGVydHMuMS53ZWlnaHQnXQ0KICAgIGRvd25bbl9yb3V0ZWRfZXhwZXJ0cyAqIGRfZXhwZXJ0OiwgOl0gPSB3ZWlnaHRzWydzaGFyZWRfZXhwZXJ0cy4yLndlaWdodCddDQoNCiAgICBmb3IgYiBpbiByYW5nZShiYXRjaF9zaXplKToNCiAgICAgICAgaW5wdXQgPSBpbnB1dF90ZW5zb3JbYl0NCiAgICAgICAgc2NvcmUgPSB0b3JjaC5ubi5mdW5jdGlvbmFsLnNvZnRtYXgoaW5wdXQgQCByb3V0ZXIsIGRpbT0xKQ0KICAgICAgICBoaWRkZW4gPSAodG9yY2gubm4uZnVuY3Rpb25hbC5zaWx1KGlucHV0IEAgZ2F0ZSkpICogKGlucHV0IEAgdXApDQogICAgICAgIGZvciBpIGluIHJhbmdlKG5fcm91dGVkX2V4cGVydHMgKyBuX3NoYXJlZF9leHBlcnRzKToNCiAgICAgICAgICAgIGRlbnNlW2IsIDosIGkgKiBkX2hpZGRlbjooaSArIDEpICogZF9oaWRkZW5dID0gaGlkZGVuWzosIGkgKiBkX2V4cGVydDooaSArIDEpICogZF9leHBlcnRdIEAgZG93bltpICogZF9leHBlcnQ6KGkgKyAxKSAqIGRfZXhwZXJ0LCA6XQ0KICAgICAgICBmb3IgdCBpbiByYW5nZShzZXFfbGVuKToNCiAgICAgICAgICAgIHRvcF92YWx1ZXMsIHRvcF9pbmRpY2VzID0gdG9yY2gudG9wayhzY29yZVt0XSwgbl9leHBlcnRzX3Blcl90b2tlbikNCiAgICAgICAgICAgIGZvciBpIGluIHJhbmdlKG5fZXhwZXJ0c19wZXJfdG9rZW4pOg0KICAgICAgICAgICAgICAgIG91dHB1dFtiLCB0XSArPSB0b3BfdmFsdWVzW2ldICogZGVuc2VbYiwgdCwgdG9wX2luZGljZXNbaV0uaXRlbSgpICogZF9oaWRkZW46KHRvcF9pbmRpY2VzW2ldLml0ZW0oKSArIDEpICogZF9oaWRkZW5dDQogICAgICAgICAgICBmb3IgaSBpbiByYW5nZShuX3NoYXJlZF9leHBlcnRzKToNCiAgICAgICAgICAgICAgICBvdXRwdXRbYiwgdF0gKz0gZGVuc2VbYiwgdCwgKG5fcm91dGVkX2V4cGVydHMgKyBpKSAqIGRfaGlkZGVuOihuX3JvdXRlZF9leHBlcnRzICsgaSArIDEpICogZF9oaWRkZW5dDQoNCiAgICByZXR1cm4gb3V0cHV0DQoNCiAgICAnJycNCiAgICBzY29yZSA9IGlucHV0IEAgcm91dGVyDQogICAgc2NvcmUgPSB0b3JjaC5ubi5mdW5jdGlvbmFsLnNvZnRtYXgoc2NvcmUsIGRpbT0xKQ0KICAgIGhpZGRlbiA9ICh0b3JjaC5ubi5mdW5jdGlvbmFsLnNpbHUoaW5wdXQgQCBnYXRlKSkgKiAoaW5wdXQgQCB1cCkNCiAgICBmb3IgaSBpbiByYW5nZShuX3JvdXRlZF9leHBlcnRzICsgbl9zaGFyZWRfZXhwZXJ0cyk6DQogICAgICAgIGRlbnNlWzosIGkgKiBkX2hpZGRlbjooaSArIDEpICogZF9oaWRkZW5dID0gaGlkZGVuWzosIGkgKiBkX2V4cGVydDooaSArIDEpICogZF9leHBlcnRdIEAgZG93bltpICogZF9leHBlcnQ6KGkgKyAxKSAqIGRfZXhwZXJ0LCA6XQ0KICAgIGZvciB0IGluIHJhbmdlKHNlcV9sZW4pOg0KICAgICAgICB0b3BfdmFsdWVzLCB0b3BfaW5kaWNlcyA9IHRvcmNoLnRvcGsoc2NvcmVbdF0sIG5fZXhwZXJ0c19wZXJfdG9rZW4pDQogICAgICAgIGZvciBpIGluIHJhbmdlKG5fZXhwZXJ0c19wZXJfdG9rZW4pOg0KICAgICAgICAgICAgb3V0cHV0W3RdICs9IHRvcF92YWx1ZXNbaV0gKiBkZW5zZVt0LCB0b3BfaW5kaWNlc1tpXS5pdGVtKCkgKiBkX2hpZGRlbjoodG9wX2luZGljZXNbaV0uaXRlbSgpICsgMSkgKiBkX2hpZGRlbl0NCiAgICAgICAgb3V0cHV0W3RdICs9IGRlbnNlW3QsIG5fcm91dGVkX2V4cGVydHMgKiBkX2hpZGRlbjpdDQoNCiAgICByZXR1cm4gb3V0cHV0LnJlc2hhcGUoMSwgc2VxX2xlbiwgZF9oaWRkZW4pDQogICAgJycn

Source code from GPU Mode and the KernelBot dataset · June 9 Researcher Reciprocity License v1.0

Best evidence level for this revision: reported

JSON