Recursive
License unknown
Use it
Benchmark submission only · no public source · license unknown
No install recipe recorded for this revision.
interfaceunknown
Compatibility
measured onNVIDIA B200
declared hardwareNVIDIA B200
architecturessm_100
dtypesbf16, fp32
Benchmark evidence
1 measurement across 1 GPU, fastest first.
Operation / workload
Hardware
Latency
Rank
Observed
Fused attention QK matmul scale mask softmax backwardsuite of 16 cases
NVIDIA B200
144.6µs
#2 of 12
2026-06-07
Reported · How evidence levels are derived →
Source and license
sourcenot public
revision digestsha256:ecd84fcaa9898a6b66832e03173fd9302f5296467b75b3a3e989e4de25fe8378
license declaredunknown
license concludedunknown
authorsRecursive
imported2026-08-24
Best evidence level for this revision: reported
JSON