Skip to content
KernelIndex
Search⌘K
Sign in

Liger

Liger-Kernel · triton · BSD-2-Clause
holds 147 records

Use it

Installable · pip · pinned to the measured build · BSD-2-ClauseView source →

pip install "liger-kernel==0.5.10"

Pinned: this command installs the newest release the evidence below measured.

interfacetriton
Compatibility
measured onNVIDIA GeForce RTX 3090, NVIDIA H100
declared hardwaredeclared only
architectures—
dtypes

Benchmark evidence

168 measurements across 2 GPUs, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA GeForce RTX 3090
155.2µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA GeForce RTX 3090
170.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA GeForce RTX 3090
170.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA GeForce RTX 3090
223.2µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA GeForce RTX 3090
236.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA GeForce RTX 3090
237.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA GeForce RTX 3090
237.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA GeForce RTX 3090
242.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA GeForce RTX 3090
246.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA GeForce RTX 3090
250.9µs
#1 of 2
2025-05-27
Run →
Show all 168 measurements ›
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA H100
256.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA H100
256.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA H100
256.9µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA H100
258.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA H100
258.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA H100
259.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA H100
259.9µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA H100
260.3µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA H100
260.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA H100
261.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA H100
263.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA H100
264.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA H100
264.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA H100
265.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA H100
315.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA H100
315.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA GeForce RTX 3090
330.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA GeForce RTX 3090
330.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA H100
353.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA H100
412.9µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA GeForce RTX 3090
431.1µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA GeForce RTX 3090
456.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA H100
491.1µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA H100
491.2µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA H100
493.5µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA H100
494.2µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA H100
495.6µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA H100
505.6µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA H100
507.3µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA H100
510.2µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA H100
510.4µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA H100
513.0µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA H100
518.7µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA H100
519.9µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA H100
520.1µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA H100
528.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA GeForce RTX 3090
546.1µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA GeForce RTX 3090
561.3µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA GeForce RTX 3090
575.7µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA GeForce RTX 3090
590.8µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA GeForce RTX 3090
592.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA GeForce RTX 3090
634.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA GeForce RTX 3090
643.1µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA GeForce RTX 3090
645.6µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA GeForce RTX 3090
648.4µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA H100
680.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA GeForce RTX 3090
687.3µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA H100
688.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA GeForce RTX 3090
692.4µs
#2 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA GeForce RTX 3090
701.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA H100
734.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA GeForce RTX 3090
785.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA H100
786.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA GeForce RTX 3090
801.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA H100
824.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA H100
826.1µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA GeForce RTX 3090
854.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA GeForce RTX 3090
855.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA H100
880.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA H100
890.8µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA H100
893.9µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA H100
894.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA H100
899.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA H100
908.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA H100
913.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA H100
914.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA GeForce RTX 3090
915.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA H100
915.6µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA H100
917.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA H100
921.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA H100
922.5µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA H100
923.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA H100
924.4µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA H100
936.0µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA H100
978.1µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA H100
995.1µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA H100
998.7µs
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA GeForce RTX 3090
1.02ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA H100
1.14ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA H100
1.29ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA GeForce RTX 3090
1.47ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA GeForce RTX 3090
1.47ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA H100
1.53ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA GeForce RTX 3090
1.55ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA H100
1.74ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA H100
1.74ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA H100
1.85ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA GeForce RTX 3090
1.95ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA GeForce RTX 3090
1.95ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA H100
1.96ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA H100
2.28ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA GeForce RTX 3090
2.37ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA GeForce RTX 3090
2.37ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=64
NVIDIA GeForce RTX 3090
2.54ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA H100
2.55ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA H100
2.57ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA H100
2.57ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA H100
2.57ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA GeForce RTX 3090
2.58ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=1
NVIDIA GeForce RTX 3090
2.60ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=256
NVIDIA GeForce RTX 3090
2.61ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=128
NVIDIA GeForce RTX 3090
2.62ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=64
NVIDIA GeForce RTX 3090
2.62ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=128
NVIDIA GeForce RTX 3090
2.64ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=1
NVIDIA GeForce RTX 3090
2.65ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA H100
2.66ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=256 · dilation=2
NVIDIA GeForce RTX 3090
2.67ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=256
NVIDIA GeForce RTX 3090
2.68ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=128 · dilation=2
NVIDIA GeForce RTX 3090
2.68ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=1
NVIDIA GeForce RTX 3090
2.72ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=512
NVIDIA GeForce RTX 3090
2.73ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=1
NVIDIA GeForce RTX 3090
2.74ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=64 · dilation=2
NVIDIA GeForce RTX 3090
2.76ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=512 · dilation=2
NVIDIA GeForce RTX 3090
2.80ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=1
NVIDIA GeForce RTX 3090
2.83ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=1024 · dilation=2
NVIDIA GeForce RTX 3090
2.88ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA H100
2.92ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=512
NVIDIA GeForce RTX 3090
2.96ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA GeForce RTX 3090
3.64ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA H100
3.79ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA GeForce RTX 3090
4.15ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA GeForce RTX 3090
4.15ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA GeForce RTX 3090
4.93ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA H100
5.20ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA H100
5.22ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA H100
5.24ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA GeForce RTX 3090
5.41ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=1024
NVIDIA GeForce RTX 3090
5.59ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA H100
5.61ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA H100
5.95ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=2
NVIDIA GeForce RTX 3090
6.52ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=2048 · dilation=1
NVIDIA GeForce RTX 3090
6.53ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=1024
NVIDIA GeForce RTX 3090
7.52ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA H100
7.70ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA H100
7.79ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA H100
7.84ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA GeForce RTX 3090
7.86ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA H100
8.18ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA GeForce RTX 3090
8.25ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA GeForce RTX 3090
8.47ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA GeForce RTX 3090
10.2ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA H100
11.6ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA GeForce RTX 3090
15.1ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA GeForce RTX 3090
15.2ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA GeForce RTX 3090
15.2ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=2048
NVIDIA GeForce RTX 3090
15.9ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA H100
16.3ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA H100
17.5ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA GeForce RTX 3090
18.6ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=2048
NVIDIA GeForce RTX 3090
23.0ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=2
NVIDIA GeForce RTX 3090
23.5ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [512] · seq=4096 · dilation=1
NVIDIA GeForce RTX 3090
23.7ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA H100
24.5ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA GeForce RTX 3090
27.1ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA GeForce RTX 3090
35.7ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA GeForce RTX 3090
52.1ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [1024] · seq=4096
NVIDIA GeForce RTX 3090
54.3ms
#1 of 2
2025-05-27
Run →
Fused neighborhood attentionfp32 · [768] · seq=4096
NVIDIA GeForce RTX 3090
79.1ms
#1 of 2
2025-05-27
Run →

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:743dfc56dca4a73f6c3b606b0cb54d12269bd3c1ce66f033d462e708716396cf
license declaredBSD-2-Clause
license concludedBSD-2-Clause
imported2026-08-20

Best evidence level for this revision: reported

JSON