Skip to content
KernelIndex
Search⌘K

PyTorch

python · BSD-3-Clause
holds 13 records

Use it

Source available upstream · BSD-3-ClauseView source →

No install recipe recorded for this revision.

interfacepython
Compatibility
measured onNVIDIA GeForce RTX 3090
declared hardwaredeclared only
architectures—
dtypes

Benchmark evidence

36 measurements across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
Multi-token attentionbf16 · [4] · seq=64
NVIDIA GeForce RTX 3090
63.5µs
#1 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=128
NVIDIA GeForce RTX 3090
90.1µs
#1 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=32
NVIDIA GeForce RTX 3090
98.3µs
#1 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=256
NVIDIA GeForce RTX 3090
163.8µs
#1 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=128
NVIDIA GeForce RTX 3090
240.8µs
#1 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=64
NVIDIA GeForce RTX 3090
241.7µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=256
NVIDIA GeForce RTX 3090
241.7µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=128
NVIDIA GeForce RTX 3090
242.7µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=32
NVIDIA GeForce RTX 3090
246.8µs
#2 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=256
NVIDIA GeForce RTX 3090
246.9µs
#1 of 2
2025-04-30
Show all 36 measurements ›
Multi-token attentionfp32 · [4] · seq=32
NVIDIA GeForce RTX 3090
258.0µs
#1 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=64
NVIDIA GeForce RTX 3090
258.0µs
#1 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=512
NVIDIA GeForce RTX 3090
313.3µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=512
NVIDIA GeForce RTX 3090
520.2µs
#2 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=128
NVIDIA GeForce RTX 3090
581.6µs
#2 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=32
NVIDIA GeForce RTX 3090
588.8µs
#2 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=64
NVIDIA GeForce RTX 3090
601.1µs
#2 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=512
NVIDIA GeForce RTX 3090
704.5µs
#2 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=1024
NVIDIA GeForce RTX 3090
719.9µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=64
NVIDIA GeForce RTX 3090
736.3µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=128
NVIDIA GeForce RTX 3090
768.0µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=256
NVIDIA GeForce RTX 3090
768.5µs
#2 of 2
2025-04-28
Multi-token attentionbf16 · [4] · seq=32
NVIDIA GeForce RTX 3090
795.2µs
#1 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=256
NVIDIA GeForce RTX 3090
859.1µs
#2 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=512
NVIDIA GeForce RTX 3090
967.7µs
#2 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=128
NVIDIA GeForce RTX 3090
1.46ms
#1 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=64
NVIDIA GeForce RTX 3090
1.49ms
#1 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=32
NVIDIA GeForce RTX 3090
1.57ms
#1 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=256
NVIDIA GeForce RTX 3090
1.74ms
#1 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=512
NVIDIA GeForce RTX 3090
1.93ms
#2 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=1024
NVIDIA GeForce RTX 3090
1.98ms
#2 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=1024
NVIDIA GeForce RTX 3090
2.70ms
#2 of 2
2025-04-30
Multi-token attentionbf16 · [4] · seq=1024
NVIDIA GeForce RTX 3090
2.77ms
#2 of 2
2025-04-28
Multi-token attentionfp32 · [4] · seq=512
NVIDIA GeForce RTX 3090
2.80ms
#2 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=1024
NVIDIA GeForce RTX 3090
6.77ms
#2 of 2
2025-04-30
Multi-token attentionfp32 · [4] · seq=1024
NVIDIA GeForce RTX 3090
9.51ms
#2 of 2
2025-04-30

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:6d4198771fde93bb65a4b8958b54d34a4cf41b41ae68b21aa09bf478b4aa6660
license declaredBSD-3-Clause
license concludedBSD-3-Clause
imported2026-08-20

Best evidence level for this revision: reported

JSON